Comment on I spy with my computer vision eye… Wally? Waldo? by Mind the Leap   
Those articles do a good job of summarising the positive side of what has been achieved (improved classification and prediction) and presenting the point that there is a lot more to intelligence than has been demonstrated so far. Though, as I mention above, a great thing about Hinton's deep learning systems is that they can generate images (and probably sounds) from incomplete or noisy input images, or by exciting high-level neurons that represent that class. Although being able to recognise images or sounds alone makes for pretty shallow "understanding". I think further advances might start to occur when they start trying to mix multiple sensory streams: particularly vision and sound, but eventually tactile senses as well. This could easily be achieved with the minimal changes to the current designs of deep learning architectures. If we can have the machine be shown an image, and it spontaneously generate speech that says what object it recognises, some people would find that a bit eerie. This would approximate playing the naming game with a child. If we could tell the robot a story, and in its "mind" it spontaneously starts generating images that represent the story, we could say that the machine is actually on the way to having a genuine "understanding" of the story. As I describe in this post, advances might also be achieved by expanding the perceptual applications to include things like motion perception: often visible as objects changing position, scale or rotation. One thing that is less obvious, is how to apply deep learning to motion control of robots. That's something I'm still thinking about. Even though I can see great potential in deep learning, I can still see some significant obstacles. Deep learning still uses iterative training techniques, which despite being quicker than they were, are still a far cry from being able to briefly see a person or object once, and then recognising that person or object from obscure angles and distances moments later. Something that people can (sometimes) do very well. Though this might be easily taken care of with some complementary memory processes (much as it seems to be in the brain). And even if current systems for deep learning turns out to have some fatal flaw in being applicable to simulating intelligence, I think the cat is very nearly out of the bag. The neocortex has largely the same structure all over the brain, and we have reasonably good approximations for many of the less regular areas of the brain. So, in my opinion it is only a matter of time before we have intelligent machines. Or machines "intelligent" enough to do most of what we've ever wanted them to, though with certain physical constraints withstanding.
          Comment on A Deep Learning Performance Lens for Low Precision Inference by OranjeeGeneral   
So how does that fit in that Baidu's Chief AI scientist has quit Baidu and opened up its own start-up company?
          Comment on The Biggest Shift in Supercomputing Since GPU Acceleration by jimmy   
@Rob, the deep learning algorithms for object recognition by far surpass anything that people were able to with classical deterministic models. That's why they are being used for self-driving cars; they have been proven to increase driver safety by a good margin. You can mention the one-off cases in the early days of self-driving, but that's not an interesting statistic at all. Deep learning is essentially an attempt to make sense of tons of noisy data, and many of the models today are not understood by their authors: "hey we did this and now we got this awesome result", very ad-hoc. In the end though, it's all statistical-mathematics, it's just that at the moment the slightly theoretically challenged CS folks are playing with this new toy, and mathematical understanding is inevitable.
          Deep Learning Segmentation of Optical Microscopy Images Improves 3-D Neuron Reconstruction   
Digital reconstruction, or tracing, of 3-D neuron structure from microscopy images is a critical step toward reversing engineering the wiring and anatomy of a brain. Despite a number of prior attempts, this task remains very challenging, especially when images are contaminated by noises or have discontinued segments of neurite patterns. An approach for addressing such problems is to identify the locations of neuronal voxels using image segmentation methods, prior to applying tracing or reconstruction techniques. This preprocessing step is expected to remove noises in the data, thereby leading to improved reconstruction results. In this paper, we proposed to use 3-D convolutional neural networks (CNNs) for segmenting the neuronal microscopy images. Specifically, we designed a novel CNN architecture, that takes volumetric images as the inputs and their voxel-wise segmentation maps as the outputs. The developed architecture allows us to train and predict using large microscopy images in an end-to-end manner. We evaluated the performance of our model on a variety of challenging 3-D microscopy images from different organisms. Results showed that the proposed methods improved the tracing performance significantly when combined with different reconstruction algorithms.
          Detection and Localization of Robotic Tools in Robot-Assisted Surgery Videos Using Deep Neural Networks for Region Proposal and Detection   
Video understanding of robot-assisted surgery (RAS) videos is an active research area. Modeling the gestures and skill level of surgeons presents an interesting problem. The insights drawn may be applied in effective skill acquisition, objective skill assessment, real-time feedback, and human–robot collaborative surgeries. We propose a solution to the tool detection and localization open problem in RAS video understanding, using a strictly computer vision approach and the recent advances of deep learning. We propose an architecture using multimodal convolutional neural networks for fast detection and localization of tools in RAS videos. To the best of our knowledge, this approach will be the first to incorporate deep neural networks for tool detection and localization in RAS videos. Our architecture applies a region proposal network (RPN) and a multimodal two stream convolutional network for object detection to jointly predict objectness and localization on a fusion of image and temporal motion cues. Our results with an average precision of 91% and a mean computation time of 0.1 s per test frame detection indicate that our study is superior to conventionally used methods for medical imaging while also emphasizing the benefits of using RPN for precision and efficiency. We also introduce a new data set, ATLAS Dione, for RAS video understanding. Our data set provides video data of ten surgeons from Roswell Park Cancer Institute, Buffalo, NY, USA, performing six different surgical tasks on the daVinci Surgical System (dVSS) with annotations of robotic tools per frame.
          A Dataset and a Technique for Generalized Nuclear Segmentation for Computational Pathology   
Nuclear segmentation in digital microscopic tissue images can enable extraction of high-quality features for nuclear morphometrics and other analysis in computational pathology. Conventional image processing techniques, such as Otsu thresholding and watershed segmentation, do not work effectively on challenging cases, such as chromatin-sparse and crowded nuclei. In contrast, machine learning-based segmentation can generalize across various nuclear appearances. However, training machine learning algorithms requires data sets of images, in which a vast number of nuclei have been annotated. Publicly accessible and annotated data sets, along with widely agreed upon metrics to compare techniques, have catalyzed tremendous innovation and progress on other image classification problems, particularly in object recognition. Inspired by their success, we introduce a large publicly accessible data set of hematoxylin and eosin (H&E)-stained tissue images with more than 21000 painstakingly annotated nuclear boundaries, whose quality was validated by a medical doctor. Because our data set is taken from multiple hospitals and includes a diversity of nuclear appearances from several patients, disease states, and organs, techniques trained on it are likely to generalize well and work right out-of-the-box on other H&E-stained images. We also propose a new metric to evaluate nuclear segmentation results that penalizes object- and pixel-level errors in a unified manner, unlike previous metrics that penalize only one type of error. We also propose a segmentation technique based on deep learning that lays a special emphasis on identifying the nuclear boundaries, including those between the touching or overlapping nuclei, and works well on a diverse set of test images.
          Baidu releases open source deep learning benchmark tool to measure inference   
Baidu releases open source deep learning benchmark tool to measure inference

Baidu Research, a division of Chinese Internet giant Baidu, has released its open source deep learning benchmark tool. Called DeepBench, the new solution comes ...

The post Baidu releases open source deep learning benchmark tool to measure inference appeared first on Open Source For You.


          NVIDIA: Startup to Poach Poachers Using Intelligent Drones   
Across the African savanna, 25,000 elephants and 1,000 rhinos are killed by poachers each year, according to some estimates. At this rate, they’ll be extinct within two decades. To combat this crisis, Neurala, a Boston-based startup, is bringing stealth, speed and scale to the fight in the form of deep learning-powered drones. By putting intelligent […]
          Microsoft made its AI work on a $10 Raspberry Pi   

When you're far from a cell tower and need to figure out if that bluebird is Sialia sialis or Sialia mexicana, no cloud server is going to help you. That's why companies are squeezing AI onto portable devices, and Microsoft has just taken that to a new extreme by putting deep learning algorithms onto a Raspberry Pi. The goals is to get AI onto "dumb" devices like sprinklers, medical implants and soil sensors to make them more useful, even if there's no supercomputer or internet connection in sight.

Via: Mashable

Source: Microsoft


          600 Dolara süper bilgisayar   

Gel vatandaş, çalan satmaz. Param olsa da ben alsam...

Şaka bir yana, teknoloji dünyası hızla yürüyor. Artık 1TFlop bir bilgisayarı 600 dolara alıp HDMI ile ekrana bağlayabilir, içine Ubuntu 14.04 kurabilirsiniz. Üstelik SSD bağlayabilir, USB 3.0 ile yardırabilir, Matlab olsun, Ansys olsun at koşturabilir, istediğiniz server işlemlerini dehşet verici güçle yapabilirsiniz.

Üstelik kamera ve I/O uyumlu, 4K video işleme kapasitesi ile geliyor. Dahası, deep learning platformu da açık kaynak olarak hazır.

MIT öğrencileri bu bilgisayarla öğrenen bir arabayı yapıp nasıl yapıldığını anlatan videoyu internete koymuşlar bile.

Boyutları kredi kartı kadar, güç tüketimi sadece 10W.

Alın size Türkçe tanıtımı;

http://www.nvidia.com.tr/object/jetson-tk1-embedde...

Artık büyük bütçelere ihtiyaç yok gençler, çalışan kafalar konuşuyor. Kafaları işletin.


          BigDL Democratizes Deep Learning Innovation   

Deep learning—a subset of machine learning and a key technique of artificial intelligence (AI)—is the most rapidly growing area of AI innovation. Big data and deep learning are technologies cut from the same cloth. They’re both enabled by the explosion of data brought about by the digital world.  But to deal effectively with mountains of ...continue reading BigDL Democratizes Deep Learning Innovation

The post BigDL Democratizes Deep Learning Innovation appeared first on IT Peer Network.

Read more >

The post BigDL Democratizes Deep Learning Innovation appeared first on Blogs@Intel.


          機械学習/Deep Learningが気になる人も要注目、「アルゴリズム」の基本が学べる無料の電子書籍150ページ   

人気過去連載を電子書籍化して無料ダウンロード提供する@IT eBookシリーズ。第29弾では「コーディングに役立つ!アルゴリズムの基本」10回分を1冊のPDFとしてまとめた。アルゴリズムとは何か? なぜ学ぶべきなのだろうか?


          BigDL Democratizes Deep Learning Innovation   

Deep learning—a subset of machine learning and a key technique of artificial intelligence (AI)—is the most rapidly growing area of AI innovation. Big data and deep learning are technologies cut from the same cloth. They’re both enabled by the explosion of data brought about by the digital world.  But to deal effectively with mountains of ...continue reading BigDL Democratizes Deep Learning Innovation

The post BigDL Democratizes Deep Learning Innovation appeared first on IT Peer Network.

Read more >

The post BigDL Democratizes Deep Learning Innovation appeared first on Blogs@Intel.


          SM Entertainment Teams Up With AI Technology Company To Create Virtual Assistants Out Of Your Favorite Celebrities   

SM Entertainment has established Artificial Intelligence (AI) technology agency AI Stars Limited with startup ObEN to bring new forms of content to fans. ObEN is an AI technology firm located in California that possesses “deep learning” technology that enables speedy creation of artificial intelligence models, such as TTS (Text-To-Speech) and avatars, with minimal data. SM Entertainment and ObEN are teaming up to […]

The post SM Entertainment Teams Up With AI Technology Company To Create Virtual Assistants Out Of Your Favorite Celebrities appeared first on Soompi.


          Maybe Trump’s behavior is explained by a simple Machine Learning (A.I.) algorithm.    
Burton offers an intriguing explanation for our inability to predict Donald Trump’s next move suggesting:
...that Trump doesn’t operate within conventional human cognitive constraints, but rather is a new life form, a rudimentary artificial intelligence-based learning machine. When we strip away all moral, ethical and ideological considerations from his decisions and see them strictly in the light of machine learning, his behavior makes perfect sense.
Consider how deep learning occurs in neural networks such as Google’s Deep Mind or IBM’s Deep Blue and Watson. In the beginning, each network analyzes a number of previously recorded games, and then, through trial and error, the network tests out various strategies. Connections for winning moves are enhanced; losing connections are pruned away. The network has no idea what it is doing or why one play is better than another. It isn’t saddled with any confounding principles such as what constitutes socially acceptable or unacceptable behavior or which decisions might result in negative downstream consequences.
Now up the stakes…ask a neural network to figure out the optimal strategy…for the United States presidency. In this hypothetical, let’s input and analyze all available written and spoken word — from mainstream media commentary to the most obscure one-off crank pamphlets. After running simulations of various hypotheses, the network will serve up its suggestions. It might show Trump which areas of the country are most likely to respond to personal appearances, which rallies and town hall meetings will generate the greatest photo op and TV coverage, and which publicly manifest personality traits will garner the most votes. If it determines that outrage is the only road to the presidency, it will tell Trump when and where his opinions must be scandalous and offensively polarizing.
Following the successful election, it chews on new data. When it recognizes that Obamacare won’t be easily repealed or replaced, that token intervention in Syria can’t be avoided, that NATO is a necessity and that pulling out of the Paris climate accord may create worldwide resentment, it has no qualms about changing policies and priorities. From an A.I. vantage point, the absence of a coherent agenda is entirely understandable. For example, a consistent long-term foreign policy requires a steadfastness contrary to a learning machine’s constant upgrading in response to new data.
As there are no lines of reasoning driving the network’s actions, it is not possible to reverse engineer the network to reveal the “why” of any decision. Asking why a network chose a particular action is like asking why Amazon might recommend James Ellroy and Elmore Leonard novels to someone who has just purchased “Crime and Punishment.” There is no underlying understanding of the nature of the books; the association is strictly a matter of analyzing Amazon’s click and purchase data. Without explanatory reasoning driving decision making, counterarguments become irrelevant.
Once we accept that Donald Trump represents a black-box, first-generation artificial-intelligence president driven solely by self-selected data and widely fluctuating criteria of success, we can get down to the really hard question confronting our collective future: Is there a way to affect changes in a machine devoid of the common features that bind humanity?

          InfiniBand Continues Momentum on Latest TOP500   

Today the InfiniBand Trade Association (IBTA) announced that InfiniBand remains the most used HPC interconnect on the TOP500. Additionally, the majority of newly listed TOP500 supercomputers are accelerated by InfiniBand technology. These results reflect continued industry demand for InfiniBand’s unparalleled combination of network bandwidth, low latency, scalability and efficiency.

As demonstrated on the June 2017 TOP500 supercomputer list, InfiniBand is the high-performance interconnect of choice for HPC and Deep Learning platforms,” said Bill Lee, IBTA Marketing Working Group Co-Chair. “The key capabilities of RDMA, software-defined architecture, and the smart accelerations that the InfiniBand providers have brought with their offering resulted in enabling world-leading performance and scalability for InfiniBand-connected supercomputers.”

The post InfiniBand Continues Momentum on Latest TOP500 appeared first on insideHPC.


          Research & Technology Manager Computer Vision and Deep Learning - Siemens - Princeton, NJ   
Define and manage execution of the Research Group’s business strategy including target definitions, growth areas and business plans....
From Siemens - Mon, 05 Jun 2017 20:52:09 GMT - View all Princeton, NJ jobs
          Artificial Intelligence (AI) Technical Sales Specialist - Intel - Santa Clara, CA   
Inside this Business Group. At least 1+ years' experience in machine learning / deep learning. Conducting deep dive technical training of Intel AI products &...
From Intel - Wed, 08 Mar 2017 11:17:48 GMT - View all Santa Clara, CA jobs
          Fontfit, ferramenta online para testar diferentes fontes em nosso site   
Frequentemente, mostramos por aqui ferramentas super úteis relacionadas ao mundo do design. Uma das coisas mais complicadas na hora de, por exemplo, criar um site, é encontrar a fonte ideal. Por isso, resolvemos apresentar Fontfit, um site que permite testar fontes em tempo real. Leia também: FONTJOY, PARA OBTER COMBINAÇÕES DE FONTES MEDIANTE DEEP LEARNING […]
          Data scientist munkakörbe keresünk munkatársat. | Feladatok: Interact with customers to underst...   
Data scientist munkakörbe keresünk munkatársat. | Feladatok: Interact with customers to understand their requirements and identify emerging opportunities. • Take part in high and detailed level solution design to propose solutions and translating them into functional and technical specifications. • Convert large volumes of structured and unstructured data using advanced analytical solutions into actionable insights and business value. • Work independently and provide guidance to less experienced colleagues/employees. • Participate in projects, closely work and collaborate effectively with onsite and offsite teams at different worldwide locations in Hungary/China/US while delivering and implementing solutions. • Continuously follow data scientist trends and related technology evolutions in order to develop knowledge base within team.. | Mit ajánlunk: To be a member of dynamically growing site and enthusiastic team. • Professional challenges and opportunities to work with prestigious multinational companies. • Competitive salary and further career opportunities. | Elvárások: Bachelor?s/Master?s Degree in Computer Science, Math, Applied Statistics or a related field. • At least 3 years of experience in modeling, segmentation, statistical analysis. • Demonstrated experience in Data Mining, Machine Learning, additionally Deep Learning Tensorflow or Natural Language Processing is an advantage. • Strong programming skills using Python, R, SQL and experience in algorithms. • Experience working on big data and related tools Hadoop, Spark • Open to improve his/her skills, competencies and learn new techniques and methodologies. • Strong analytical and problem solving skills to identify and resolve issues proactively • Ability to work and cooperate onsite and offsite teams located in different countries Hungary, China, US and time zones. • Strong verbal and written English communication skills • Ability to handle strict deadlines and multiple tasks. | További infó és jelentkezés itt: www.profession.hu/allas/1033284
          Research & Technology Manager Computer Vision and Deep Learning - Siemens - Princeton, NJ   
Define and manage execution of the Research Group’s business strategy including target definitions, growth areas and business plans....
From Siemens - Mon, 05 Jun 2017 20:52:09 GMT - View all Princeton, NJ jobs
          Artificial Intelligence (AI) Technical Sales Specialist - Intel - Santa Clara, CA   
Inside this Business Group. At least 1+ years' experience in machine learning / deep learning. Conducting deep dive technical training of Intel AI products &...
From Intel - Wed, 08 Mar 2017 11:17:48 GMT - View all Santa Clara, CA jobs
          李彦宏:Facebook的新使命,有点像百度贴吧   

6月22日,扎克伯格在芝加哥参加Facebook的活动的时候,宣布改了使命,过去的使命是“Make the world more open and connected”,新的使命是“Bring the world closer together”。对此,6月29日,百度创始人、董事长李彦宏在首届世界智能大会上说,“Facebook的新使命,说起来有点像贴吧。”

李彦宏说,Facebook过去的使命是非常具有互联网时代特色的,就是开放、连接。现在扎克伯格也意识到这个东西不够了,他说要“Bring the world closer together”,怎么才能“closer together”呢?还是要通过用户画像、通过人工智能的技术,找到人和人之间相同的兴趣,把他们连接在一起。

两个人相隔千里,如果大家都对牡丹花感兴趣,他们靠用户画像可以把彼此匹配起来。牡丹花有一千多个品种,有姚黄魏紫之类的各种各样,然而当你说这样一个话的时候,周围没有一个人听得懂,但是千里之外可能有另外一个人他也感兴趣这个东西,他能听得懂,两个人就可以连接起,come closer。说起来有点像贴吧,但这确实是Facebook的新使命。

此前,有媒体发表文章称贴吧要被关掉。

6月27日晚间,百度集团总裁兼首席运营官陆奇在公司内网发帖,称“贴吧要被关掉,这完全违背事实”。

陆奇说,贴吧在百度 “夯实移动基础,决胜AI时代”的一盘棋战略中有不可替代的重要地位,是百度构建内容生态的锁定型产品。

“事实上,在周、月例会及总监会上,我都明确强调了贴吧的战略意义。”陆奇说,“另外,贴吧近期发布的新版APP,也让我看到贴吧的用户体验在不断的改进和提升。未来,随着战略执行的深入,贴吧的价值将愈加凸显。希望贴吧同学们和我一起努力,为用户带来更好产品体验,我们共同来推进一盘棋战略的快速落地。”

6月27日,李彦宏也现身百度贴吧“李彦宏吧”,晒出了自己和妻子马东敏的合影,还说:“二十年前那个夏天,从这里出发去了硅谷,自此踏入互联网,现在该进入人工智能时代了。猜猜“这里”是哪里?”

QQ截图20170630145433

李彦宏在演讲时,还评价了马云、马化腾。

他说,很多人讲数据秒杀算法,马云讲了DT时代好多年了,上个月我们在大数据峰会上(贵州)也碰到,而且特别巧的是我们俩是一个平行论坛。我当时觉得压力很大,大家都去听马云讲不听我的怎么办?但是还好,我那场也挺满的。我讲的是真正推动社会进步的是算法。

到最后闭幕的时候,马化腾出来总结说其实他们俩说的都不对,最重要的既不是数据,也不是算法,是场景。其实没有必要去做这样的争论,甚至我觉得这些都是套路,大家说的本不矛盾。

我在当时讲话时,除了讲算法推动社会进步,我也讲了过去的创新来自于大学、实验室,未来的创新会来自于数据和场景。

以下是李彦宏演讲全文实录:

大家早上好,非常荣幸能够在世界智能大会上跟大家分享一些我的观点。感谢进鹏书记的介绍,过去十年我们经常在一起探讨各种各样关于人工智能方面的问题,他本人也是人工智能方面的专家,我们有非常多的共同语言。

首先,我要祝贺天津,在人类刚刚进入智能时代的时候,恰逢其时地举办了这么一个世界智能大会,能够把兴趣相同的人聚集在一起,大家一起探讨。不管是智能科技给人们未来带来的收益,还是将来我们有可能面临的风险,这些都是非常有意义的话题,所以我们在这个时候一起探讨是非常好的。

我和刚才几位发言嘉宾一样,我们都认为人工智能时代已经到来了,未来的几十年、可能三十年到五十年,推动世界经济发展的最重要力量很可能就是智能科技的进步。

为什么这样说呢?我们看一看历史,也基本上会得出来同样的结论:过去一百年,世界经济的成长主要是靠技术革新、靠创新来推动的,不是靠人口增长来推动的。过去一百年,大多数的经济增长总量是发达国家创造的,发达国家的经济增长不是人口增长带来的,而是人们生产效率提升带来的。如果我们看一下过去四十年,很明显的一个特点就是技术革新主要发生在IT领域。我可以讲,过去四十年世界经济增长的主要推动力是IT技术的革新。

怎么来证明它呢?大家来看一下,1977年的时候,美国股市上最大的五个公司主要是汽车、能源领域的公司,只有一个跟IT相关的是IBM。到2017年,现在前五家公司已经全部都是IT领域的公司。

为什么是这样的?因为过去这么多年,无数的IT方面的进步在不断地推进着世界经济的发展。现在,中国的阿里、腾讯也是逐步地在进入这样的状态。

我作为一个技术出身的人主要想看一下,如果我们把时间再拉近一点,过去二十年,主要的IT领域创新发生在什么地方?可能有人听过一个词叫“去IOE”,我最早听到这个词的时候不是很理解,什么叫“去IOE”?就是去IBM、Oracle、EMC。像我们这种互联网公司、搜索引擎公司,从成立的第一天起就没有用过IOE。最早的时候用户量很大,大家都有需求要上网找东西,但是我们不能买IBM那么贵的服务器,我们也不能买Oracle那么贵、而且还慢的数据库管理系统。

因为我们要处理的是非结构化数据,所以就重新开发了一套处理非结构化数据的技术,用很便宜的PC服务器,用自己开发的软件程序,就能够同时支持几千万人、几亿人进行搜索。

所以“去IOE”在搜索引擎公司一开始的时候就解决了这个问题。再到后来,搜索技术不仅可以在搜索里用,也可以在很多应用方面去用,怎么样大规模地并行化地处理非结构化数据?把这些原来只为搜索准备的技术generalise(普及)之后,后来我们在其他地方也可用。generalise(普及)之后叫什么呢?就是云计算。大家知道云计算这个概念是谷歌提出来的,他们公布说云计算基础是GFS、BigTable、MapReduce这套东西。

开始进入人工智能时代以后,又是搜索引擎公司最先利用原有的基础再接着创新,更往前推进一步,不仅是软件层面的创新,也到了硬件层面。大家知道GPU原来是用来玩游戏的,现在人工智能的深度学习计算基本上都是GPU来完成的。

吴恩达以前在谷歌的时候,据说很不爽,因为谷歌不让他买GPU,他们不相信这个方向。后来他到百度之后可以随便买,结果我们就有了全世界最大的GPU的人工智能深度学习计算系统。再往后FPGA也是新一代的系统架构,它可以更便宜更flexible(灵活地)去解决相关的问题。这就说明了为什么百度网盘还在支撑着,而其他的网盘都撑不住了,因为我们所用的架构成本更低。

过去二十年,搜索引擎公司在技术方面对计算机科学有着相当大的贡献,无论是“去IOE”、云计算,还是FPGA、GPU的广泛使用,都跟搜索引擎公司首先面临这个问题,并且解决它是分不开的。

去年我在百度的联盟峰会上讲“下一幕是人工智能”,其实仅仅过了一年时间就不是“下一幕”了,而是“这一幕”。现在所有的人都意识到了,我们处在人工智能时代。

人工智能时代有新的东西在不断出现,每当我们看到这些东西的时候都觉得非常兴奋。2009年,谷歌开始了他们的自动驾驶项目,大家知道今天全球无论是互联网公司,还是汽车厂商都已经意识到自动驾驶代表着汽车工业的未来。

2013年的1月份,百度成立了IDL(Institute of Deep Learning 百度深度学习实验室),这是全球第一家以深度学习命名的企业研究院。

今天,任何一个会一点深度学习技术的人,别的不敢说,找一个高薪的工作是没问题的。2014年12月份亚马逊开始内测echo,这个东西非常有划时代的意义。过去我们都在用手机,今天如果用echo这样的东西,它可以远场地进行语音识别。PC时代大家是用鼠标、键盘来和计算机交互,智能手机时代我们用触摸屏来和计算机交互,那么在人工智能时代很可能用语音和图像来和计算机进行交互。

2015年12月,百度宣布我们语音识别技术的精准度已经超越了人类人工识别。

2016年、2017年,微软和谷歌也分别宣布他们的语音识别精准度超越了人类水平。如果更近一点,我们看到2016年12月5日亚马逊推出了无人值守的线下零售商店Amazon Go,以后你进商店不管是挑东西还是付款都不需要人工操作,完全靠机器可以解决。

5月4日,我们宣布改变了百度的使命。百度成立快18年的时间,前17年的时候,我们是说“让人们最平等便捷地获取信息,找到所求”,大家能感觉到这是带有互联网特色的使命,它是连接人和信息。但是随着人工智能时代的到来,我们觉得我们能做的事情远远不只这些,智能技术可以改变更多,可以让复杂的世界变得更简单,所以我们就说是“用科技让复杂的世界更简单”。

为什么我们说这个世界还是复杂的呢?

你现在去机场还得要记着带身份证、过安检,为什么不能不带身份证不过安检直接刷脸就上飞机呢?我们每天要出行,很多人都选择开车,但开车还要学习,有的人要花一个月的时间甚至更长的时间,再花几千块钱上驾校学习才能学会开车,为什么不能我坐到车里想去哪它就帮我开到哪呢?

这样的事情,我们觉得智能技术可以使世界变得更加简单。包括我们天天在使用的电视机摇控器,有几十个按纽,大多数人其实从来不用那些按钮,也不知道那些按钮是干吗用的,为什么不能我说调到天津卫视它就调到天津卫视,为什么不能我说让它音量大点就大点,我问说那个女演员叫什么它就告诉我叫什么,这完全可以用语音技术来解决。世界将来会变得这么简单,而它变简单的途径就是靠智能技术。

上个星期,Facebook也正好宣布改了使命,6月22日,扎克伯格在芝加哥参加Facebook的活动的时候,他说我们过去的使命是“Make the world more open and connected”,这个话也是非常具有互联网时代特色的,就是开放、连接。现在他也意识到这个东西不够了,他说我们要“Bring the world closer together”,怎么才能“closer together”呢?还是要通过用户画像、通过人工智能的技术,找到人和人之间相同的兴趣,把他们连接在一起。

两个人相隔千里,如果大家都对牡丹花感兴趣,他们靠用户画像可以把彼此匹配起来。牡丹花有一千多个品种,有姚黄魏紫之类的各种各样,然而当你说这样一个话的时候,周围没有一个人听得懂,但是千里之外可能有另外一个人他也感兴趣这个东西,他能听得懂,两个人就可以连接起来、become closer。说起来有点像贴吧,但这确实是Facebook的新使命。

重视人工智能已经成为全球的共识。过去这一年,如果你和世界任何一个国家的领袖交谈,讲起人工智能的话题,他都可以跟你讲几句,因为人工智能对人类社会的影响不仅在经济层面,在政治、文化各个层面都有非常大的影响力。我们看到去年全球科技巨头在AI的投资有300亿美元,人们对AI的关注也是前所未有的高。我们看到百度搜索“人工智能”这个词的媒体指数, 2016年比2015年上升了632%,2017年上半年在这么高的基础上又上升了45%。

中国在人工智能方面还是非常有优势,我们有很大的市场,我们有很多的人才,大家看人工智能方面的论文,好多都是中国人写的,我们可能天生就适合干这个事,我们也有很多的资金。更重要的是人工智能技术要想往前推进的话,需要有大量的数据积累进行训练,全世界没有一个市场是有七亿多的网民,说的是同样的语言,他们遵循的是同样的文化和道德标准,遵循的是同样的法律。你再也找不到这样一个市场,在这样的市场当中,你在人工智能方面真的是如鱼得水。我们不领先世界,真的是说不过去的。

所以外媒也注意到了,《华盛顿邮报》说“中国已在人工智能研究方面领先美国”,《纽约时报》讲“中国正在人工智能领域超越美国”,虽然媒体的话总有点语不惊人死不休,他们的话总是说的更极端一些,但是中国在人工智能方面的成就应该说是举世瞩目的。

人工智能在应用方面,每天都有新的东西出现,正在加速进入各种各样的场景。

比如说人脸识别,就是在昨天,在南阳机场已经实现了刷脸登机,不用登机牌人就可以直接过去。这样在过去我们很难想象的事情已经实实在在地发生了。

而无人驾驶呢?全球的共识大概是在2021到2022年之间无人驾驶会成为现实。目前互联网公司、汽车运营公司以及汽车制造公司、汽车零件公司都已经加入到了无人驾驶的研究和开发行列,这个大潮是谁都挡不住的。

我们也尝试用AI的技术、用人脸识别技术去帮助寻找走失的亲人,最近的案例是重庆一个孩子在四五岁的时候走失,二十七年以后他生活在福建,通过人脸识别的比对找到了他的亲人。还有一例是陕西的一对老夫妇带着他智力障碍的儿子到北京看病,儿子走丢了,因为有智力障碍,说不出来自己姓什么叫什么。

他们在北京徘徊八个月之后,还是靠我们人脸识别的技术给他比对上了,找到他的时候,他已经满脸络腮胡了,跟平时见到的本人照片已经完全不一样了,但是机器可以识别出来。

智能语音的交互,像对电视说话、用语音来控制电视现在也已经实现了,在长沙就已经落地了,有这样的机顶盒它完全可以用语音来进行操控。你问它这个演员是谁,它真的知道是谁。

我最近也在讲人工智能时代的思考方式跟互联网时代是非常不一样的,我总结了一些。

第一,我们觉得智能手机已经完全普及。手机仍然会长期存在,但是移动互联网的机会已经不多了,如果今天你再重新创业,干一个什么事,想要靠移动互联网起来,这已经非常困难了。

第二,需要把思维方式从Think Mobile变成Think AI。Mobile时代是什么样的思维方式呢?就是什么东西都要用手滑来滑去,在设计的时候很关注这个字体大小是什么样的,那个导航要放在什么位置,是纯软件的东西,但是进入人工智能时代,你必须要思考软硬件的结合,我们在公司内部开会的时候就很明显,做移动产品的产品经理就很关注这个功能要用几个字来描述,字体大小应该是什么样的,什么颜色;而人工智能方面的产品经理从兜里掏出一个芯片说,我这个芯片现在可以做到58块钱一片,这里有什么功能。这是两种完全不同的思维方式。

所以我们也觉得,第三,未来在智能时代,软件和硬件的结合会越来越明显。

第四,很多人讲数据秒杀算法,马云讲了DT时代好多年了,上个月我们在大数据峰会上也碰到,而且特别巧的是我们俩是一个平行论坛。我当时觉得压力很大,大家都去听马云讲不听我的怎么办?但是还好,我那场也挺满的。我讲的是真正推动社会进步的是算法。到最后闭幕的时候,马化腾出来总结说其实他们俩说的都不对,最重要的既不是数据,也不是算法,是场景。其实没有必要去做这样的争论,甚至我觉得这些都是套路,大家说的本不矛盾。我在当时讲话时,除了讲算法推动社会进步,我也讲了过去的创新来自于大学、实验室,未来的创新会来自于数据和场景。

第五,因为我们已经进入了AI的时代、智能的时代,用AI的思维做互联网产品会有非常大的优势。

所以,我在这里希望大家能够把握人工智能的广阔前景,因为它确实是可以改变任何一个行业,它会改变医疗健康行业、教育行业,刚才联合国教科文组织知识社会局局长也讲到,有很多知识我们积累起来了,但怎么让人能够更加高效地、更加简单地学会这些知识?人工智能在个性化学习教育上会有很大的潜力,金融方面现在也广泛地运用人工智能技术。几乎每一个行业,比如制造、安防等都会受到人工智能技术非常大的影响。今天下午的很多论坛都是一个一个行业地在探讨人工智能的影响力,所以我就不在这里细讲了。

我想讲的是人工智能时代不是某一家公司,或者某几家公司的专利,相反它是很多公司合在一起来做的事情。最近我看到李开复在《纽约时报》上撰文,他说未来人工智能时代很残酷,最后机会都会变成大公司的。全球可能有七家公司,真正能够从人工智能时代获益变成很大的公司。

其实我不同意这个观点,要想做成一件事情,比如说现在百度在无人驾驶上做了好几年,积累了很多的技术,但是我们仍然觉得,靠我们一家公司做这个事是做不成的,我们需要把像大陆、博世这种等Tier 1(一级供应商)的厂商引进来,也要把汽车制造商引进来,也需要把共享出行的运营商引进来。

在芯片的层面,我们和intel(英特尔)合作,我们跟NVIDIA(英伟达)这样完全靠GPU起来的一个公司也要有合作,跟HTC等等很多公司合作。大家一起合作才能把智能技术推向一个新的高潮。


          VW Announces Deep Learning Partnership With California Firm   
News
This week VW announced a strategic partnership with a Silicon Valley tech firm in hopes of bolstering its artificial intelligence capabilities.
Staff Author: 
Topics: 

          InfiniBand Continues Momentum on Latest TOP500   

Today the InfiniBand Trade Association (IBTA) announced that InfiniBand remains the most used HPC interconnect on the TOP500. Additionally, the majority of newly listed TOP500 supercomputers are accelerated by InfiniBand technology. These results reflect continued industry demand for InfiniBand’s unparalleled combination of network bandwidth, low latency, scalability and efficiency.

As demonstrated on the June 2017 TOP500 supercomputer list, InfiniBand is the high-performance interconnect of choice for HPC and Deep Learning platforms,” said Bill Lee, IBTA Marketing Working Group Co-Chair. “The key capabilities of RDMA, software-defined architecture, and the smart accelerations that the InfiniBand providers have brought with their offering resulted in enabling world-leading performance and scalability for InfiniBand-connected supercomputers.”

The post InfiniBand Continues Momentum on Latest TOP500 appeared first on insideHPC.


          Google Cloud TPU (Google io 2017)   
Google stellt die zweite Generation seiner TPU vor. Die Tensor Processing Units beschleunigen jetzt auch das Training beim Deep Learning.
          Radeon Instinct (Trailer)   
AMD präsentiert seine Vision von Radeon Instinct für Deep Learning.
          Crazy Stone Deep Learning -The First Edition   
Crazy Stone nutzt Deep Neural Networks für die Computerspieler-KI.
          【NVIDIA-上海】招聘深度学习软件测试工程师   
大家好, 我是NVIDIA的HR,Carra,目前我司正在大量招聘软件测试方面的候选人,会去做CUDA,Deep Learning 以及游戏相关的项目。我们的软件测试不仅仅是测试,还需要C/C++以及编译方面的技能。如果你感兴趣的话,欢迎把简历投至我的邮箱:carraz@nvidia.com 1. Senior ...
          【NVIDIA-上海】招聘深度学习软件测试工程师   
大家好,[/backcolor] 我是NVIDIA的HR,Carra,目前我司正在大量招聘软件测试方面的候选人,会去做CUDA,Deep Learning 以及游戏相关的项目。我们的软件测试不仅仅是测试,还需要C/C++以及编译方面的技能。如果你感兴趣的话,欢迎把简历投至我的邮箱:[/backcolor] 1. ...
          Disney Research, Pixar Animation Studios and UCSB accelerate rendering with AI   
Researchers from Disney Research, Pixar Animation Studios, and the University of California, Santa Barbara have developed a new technology based on artificial intelligence (AI) and deep learning that eliminates noise from the simulation of light flow in 3D scenes and thereby enables production-quality rendering at much faster speeds.
          Tensorflow Programs and Tutorials   
This repository did some toy experiments based on Tensorflow in order to introduce some deep learning concepts which are used for image recognition and language modeling.
          Slides: Machine Learning Summer School @ Max Planck Institute for Intelligent Systems, Tübingen, Germany   


Here are the slides of some of the presentations at the Machine Learning Summer School at the Max Planck Institute for Intelligent Systems, Tübingen, Germany

Shai Ben-David
(Waterloo)
 Learning Theory.
Slides part 1 part 2 part 3
Dominik Janzing
(MPI for Intelligent Systems)
 Causality.
Slides here.
Stefanie Jegelka
(MIT)
 Submodularity.
Slides here.


Jure Lescovec
(Stanford)
Network Analysis.
Slides 1 2 3 4




Ruslan Salakhutdinov
(CMU)
Deep Learning.
Slides part 1 part 2


Suvrit Sra
(MIT)
Optimization.
Slides 1 2 3A 3B
Bharath Sriperumbudur
(PennState)
Kernel Methods.
Slides part 1 part 2 part 3


Max Welling
(Amsterdam)
Large Scale Bayesian Inference with an Application to Bayesian Deep Learning
Slides here.




Bernhard Schölkopf
(MPI for Intelligent Systems)
Introduction to ML and speak on Causality.
Slides here.

h/t Russ




Join the CompressiveSensing subreddit or the Google+ Community or the Facebook page and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

          Deep learning on Apache Spark and Apache Hadoop with Deeplearning4j   
爱可可-爱生活   网页版 2017-06-29 19:55 架构 经验总结 深度学习 Hadoop Nish […]
          Principal Engineer -- Deep Learning - Intel - Santa Clara, CA   
Network design - compression - network sparsity - attention / selective executionalso:. Principle Engineer is a senior technical position at Intel.This position...
From Intel - Fri, 12 May 2017 10:31:23 GMT - View all Santa Clara, CA jobs
          A Deep Learning Performance Lens for Low Precision Inference   

Few companies have provided better insight into how they think about new hardware for large-scale deep learning than Chinese search giant, Baidu.

As we have detailed in the past, the company’s Silicon Valley Research Lab (SVAIL) in particular has been at the cutting edge of model development and hardware experimentation, some of which is evidenced in their publicly available (and open source) DeepBench deep learning benchmarking effort, which allowed users to test different kernels across various hardware devices for training.

Today, Baidu SVAIL extended DeepBench to include support for inference as well as expanded training kernels. Also of

A Deep Learning Performance Lens for Low Precision Inference was written by Nicole Hemsoth at The Next Platform.


          The application of deep convolutional neural networks to ultrasound for modelling of dynamic states within human skeletal muscle. (arXiv:1706.09450v1 [cs.CV])   

Authors: Ryan J. Cunningham, Peter J. Harding, Ian D. Loram

This paper concerns the fully automatic direct in vivo measurement of active and passive dynamic skeletal muscle states using ultrasound imaging. Despite the long standing medical need (myopathies, neuropathies, pain, injury, ageing), currently technology (electromyography, dynamometry, shear wave imaging) provides no general, non-invasive method for online estimation of skeletal intramuscular states. Ultrasound provides a technology in which static and dynamic muscle states can be observed non-invasively, yet current computational image understanding approaches are inadequate. We propose a new approach in which deep learning methods are used for understanding the content of ultrasound images of muscle in terms of its measured state. Ultrasound data synchronized with electromyography of the calf muscles, with measures of joint torque/angle were recorded from 19 healthy participants (6 female, ages: 30 +- 7.7). A segmentation algorithm previously developed by our group was applied to extract a region of interest of the medial gastrocnemius. Then a deep convolutional neural network was trained to predict the measured states (joint angle/torque, electromyography) directly from the segmented images. Results revealed for the first time that active and passive muscle states can be measured directly from standard b-mode ultrasound images, accurately predicting for a held out test participant changes in the joint angle, electromyography, and torque with as little error as 0.022{\deg}, 0.0001V, 0.256Nm (root mean square error) respectively.


          Chord Label Personalization through Deep Learning of Integrated Harmonic Interval-based Representations. (arXiv:1706.09552v1 [cs.SD])   

Authors: H.V. Koops, W.B. de Haas, J. Bransen, A. Volk

The increasing accuracy of automatic chord estimation systems, the availability of vast amounts of heterogeneous reference annotations, and insights from annotator subjectivity research make chord label personalization increasingly important. Nevertheless, automatic chord estimation systems are historically exclusively trained and evaluated on a single reference annotation. We introduce a first approach to automatic chord label personalization by modeling subjectivity through deep learning of a harmonic interval-based chord label representation. After integrating these representations from multiple annotators, we can accurately personalize chord labels for individual annotators from a single model and the annotators' chord label vocabulary. Furthermore, we show that chord personalization using multiple reference annotations outperforms using a single reference annotation.


          Machine listening intelligence. (arXiv:1706.09557v1 [cs.SD])   

Authors: C.E. Cella

This manifesto paper will introduce machine listening intelligence, an integrated research framework for acoustic and musical signals modelling, based on signal processing, deep learning and computational musicology.


          Recurrent neural networks with specialized word embeddings for health-domain named-entity recognition. (arXiv:1706.09569v1 [cs.CL])   

Authors: Inigo Jauregi Unanue, Ehsan Zare Borzeshi, Massimo Piccardi

Background. Previous state-of-the-art systems on Drug Name Recognition (DNR) and Clinical Concept Extraction (CCE) have focused on a combination of text "feature engineering" and conventional machine learning algorithms such as conditional random fields and support vector machines. However, developing good features is inherently heavily time-consuming. Conversely, more modern machine learning approaches such as recurrent neural networks (RNNs) have proved capable of automatically learning effective features from either random assignments or automated word "embeddings". Objectives. (i) To create a highly accurate DNR and CCE system that avoids conventional, time-consuming feature engineering. (ii) To create richer, more specialized word embeddings by using health domain datasets such as MIMIC-III. (iii) To evaluate our systems over three contemporary datasets. Methods. Two deep learning methods, namely the Bidirectional LSTM and the Bidirectional LSTM-CRF, are evaluated. A CRF model is set as the baseline to compare the deep learning systems to a traditional machine learning approach. The same features are used for all the models. Results. We have obtained the best results with the Bidirectional LSTM-CRF model, which has outperformed all previously proposed systems. The specialized embeddings have helped to cover unusual words in DDI-DrugBank and DDI-MedLine, but not in the 2010 i2b2/VA IRB Revision dataset. Conclusion. We present a state-of-the-art system for DNR and CCE. Automated word embeddings has allowed us to avoid costly feature engineering and achieve higher accuracy. Nevertheless, the embeddings need to be retrained over datasets that are adequate for the domain, in order to adequately cover the domain-specific vocabulary.


          CS591 Report: Application of siamesa network in 2D transformation. (arXiv:1706.09598v1 [cs.CV])   

Authors: Dorothy Chang

Deep learning has been extensively used various aspects of computer vision area. Deep learning separate itself from traditional neural network by having a much deeper and complicated network layers in its network structures. Traditionally, deep neural network is abundantly used in computer vision tasks including classification and detection and has achieve remarkable success and set up a new state of the art results in these fields. Instead of using neural network for vision recognition and detection. I will show the ability of neural network to do image registration, synthesis of images and image retrieval in this report.


          Deep learning bank distress from news and numerical financial data. (arXiv:1706.09627v1 [stat.ML])   

Authors: Paola Cerchiello, Giancarlo Nicola, Samuel Ronnqvist, Peter Sarlin

In this paper we focus our attention on the exploitation of the information contained in financial news to enhance the performance of a classifier of bank distress. Such information should be analyzed and inserted into the predictive model in the most efficient way and this task deals with all the issues related to text analysis and specifically analysis of news media. Among the different models proposed for such purpose, we investigate one of the possible deep learning approaches, based on a doc2vec representation of the textual data, a kind of neural network able to map the sequential and symbolic text input onto a reduced latent semantic space. Afterwards, a second supervised neural network is trained combining news data with standard financial figures to classify banks whether in distressed or tranquil states, based on a small set of known distress events. Then the final aim is not only the improvement of the predictive performance of the classifier but also to assess the importance of news data in the classification process. Does news data really bring more useful information not contained in standard financial variables? Our results seem to confirm such hypothesis.


          Progress Estimation and Phase Detection for Sequential Processes. (arXiv:1702.08623v2 [cs.LG] UPDATED)   

Authors: Xinyu Li, Yanyi Zhang, Jianyu Zhang, Yueyang Chen, Shuhong Chen, Yue Gu, Moliang Zhou, Richard A. Farneth, Ivan Marsic, Randall S. Burd

Process modeling and understanding are fundamental for advanced human-computer interfaces and automation systems. Much recent research focused on activity recognition and little has been done on process progress detection from sensor data. We introduced a real-time, sensor-based system for modeling, recognizing and estimating the progress of a work process. We implemented a multimodal deep learning structure to extract the relevant spatio-temporal features from multimodal sensory inputs and used a novel deep regression structure for overall completeness estimation. Using process completeness estimation with a Gaussian mixture model, our system can predict the phase for sequential processes. The performance speed, calculated from completeness estimates, allows online estimation of the remaining time. To help the training of our system, we introduced a novel rectified hyperbolic tangent (rtanh) activation function and conditional loss. Our system was tested on data obtained from a medical process (trauma resuscitation) and sports events (Olympic swimming competition). Our system outperformed existing trauma-resuscitation phase detectors with over 86% phase detection accuracy, 0.67 F1-score, under 12.65% completeness estimation error, and less than 7.5 minutes time-remaining estimation error. For the Olympic swimming dataset, our system achieved 88% accuracy, 0.58 F1-score, 6.32% completeness estimation error and an average 2.9 minute time-remaining estimation error.


          Comment on Unfamiliarity: A Field Mark by Patrick Millar   
Regarding how modern "machine intelligence" or "deep learning" algorithms would learn to distinguish calls, it is through training. Specifically, supervised training. i.e. the algorithm is fed a collection of recordings of known songs and calls for a particular species. If (big "if") the data is comprehensive and representative of the full range of vocalizations then the algorithms (once trained) can achieve high accuracy against new calls. The more good data that can be fed to these algorithms, the better. Time of day, date, habitat call is recorded in, GPS location - all will provide clues. As for the call itself, it is ideal if the call / song is isolated from any longer recording - but these algorithms can do surprisingly well even if it is not (although the size of the training data set would have to be larger). Typically a frequency analysis would be run on the audio. All of these features are then usually put in a 1 dimensional vector that is used to train the neural net. How this is done deterministically is unclear. Neural nets in particular are still black box in terms of how they work, even to researchers. But work they do. The features selected by these neural nets as the key areas of interest / differentiators sometimes correspond to what we humans might identify, but often do not. It's intriguing. All of this to say - yes, it can be done. It is being done to a limited extent today, but you can expect excellent automatic bird call ID sooner rather than later. It will do wonders for ornithology ...
          Lucid Planet Radio with Dr. Kelly: Encore: The Singularity is HERE: Interface the Future, Explore Paradox, and Recode Mythology with Mitch Schultz    
GuestToday visionary thinker, futurist and filmmaker Mitch Schultz joins Dr. Kelly to explore humanity as we approach the technological singularity. What is the singularity, and what does it mean for humanity?   Explore a transdisciplinary approach at the intersection of the arts, cognitive psychology, deep learning, and philosophy. Guided by Consciousness, Evolution, and Story. Beginning with what conscious state of being (terrestrial and universal) is perceiving. Followed the one consta ...
          Lucid Planet Radio with Dr. Kelly: The Singularity is HERE: Interface the Future, Explore Paradox, and Recode Mythology with Mitch Schultz    
EpisodeToday visionary thinker, futurist and filmmaker Mitch Schultz joins Dr. Kelly to explore humanity as we approach the technological singularity. What is the singularity, and what does it mean for humanity?   Explore a transdisciplinary approach at the intersection of the arts, cognitive psychology, deep learning, and philosophy. Guided by Consciousness, Evolution, and Story. Beginning with what conscious state of being (terrestrial and universal) is perceiving. Followed the one consta ...
          Artificial intelligence – it’s deep (learning)    
Maschinenmensch (machine-human) on display at the Science Museum at Preview Of The Science Museum's Robots Exhibition

Maschinenmensch (machine-human) on display at the Science Museum at Preview Of The Science Museum's Robots Exhibition at Science Museum London.; Credit: Ming Yeung/Getty Images Entertainment Video

AirTalk®

To what extent can we trust artificial intelligence if we don’t understand its decision-making process?

This may sound like a science fiction scenario, but it’s an ethical dilemma that we’re already grappling with.

In his recent MIT Technology Review cover story, “The Dark Secret at the Heart of AI,” Will Knight explores the ethical problems presented by deep learning.  

Some context: one of the most efficient types of artificial intelligence is machine learning – that’s when you program a computer to write its own algorithms. Deep learning is a subset of machine learning which involves training a neural network, a mathematical approximation of the way neurons process information, often by feeding it examples and allowing it to “learn.”

This technique has taken the tech world by storm, and is already being used for language translation, image captioning and translation. The possibilities are extensive, with powerful decision making potential that can be used for self-driving cars, the military and medicine.

However, when an AI writes its own algorithm, it often becomes so complicated that a human can’t decipher it, creating a “black box,” and an ethical dilemma. What are the trade-offs to using this powerful technique? To what extent can humans trust a decision-making process that they can’t understand? How can we regulate it?

Guest host Libby Denkmann in for Larry Mantle

Guest:

Will Knight, senior editor for AI at MIT Technology Review; he wrote the article “The Dark Secret at the Heart of AI;” he tweets @willknight

This content is from Southern California Public Radio. View the original story at SCPR.org.


          Gracies Backyard - Questions Answered   
I was very impressed with Richard Perkins yesterday, after I posted my review/thoughts on the film of his farm he sent a fair bit of time and answered my questions in a very thorough way. I decided that I would post it here so that anyone who had the same questions about interns working on his farm could see his response. 


"Hi Kevin,

Thanks for your interest and support. These are important and complex questions, but I can speak to them a little here as it comes out naturally….


I have made videos about the numbers/ human-hours of our farm in the past, which you can see on Youtube. Current rates of production represent 4 full time positions. The farm pays 4 full time wages. We are heavily focused on edu work, as there are very few places like this where people can immerse themselves in scaling up permaculture to the farm, in a supportive learning environment with time, space and support to design their own properties and all the back end business planning to get started on the right foot. This is evidenced to me in that our Internship program, running right now, was quadruple booked this year. People are seeking longer term mentorship and practical experience in regenerative ag and don’t have many options, certainly here in Europe. We always have a bigger core team than needed to allow the space for deep learning to evolve for people, and judging by what some folks get up to leaving here, I see it’s going very well. We want to have a lot of fun together, as well as work hard and efficiently. We’ve become a bit of a springboard/ ‘reality check’ for folks just about to leap into making a career in regenerative enterprise, and this place/ context/ environment excels in that (the general consensus of folks here). For context, I’ve been engaged in intensive education for a long time and whilst I could, I wouldn’t just move to Sweden to farm poultry and grow veg. I’m interested in the whole we are creating here, which is what (in some ways) I wish I could have found on my learning journey.

Now, we do a bunch more than we would if we were just 4 peop’s working the farm, because we can. If it was the case of just employees, we might not have pigs, would not have cows, wouldn’t do things like plant hops, make free youtube videos, collect data, write books to try help more people,etc, etc. We are trying to use the additional brains and skills to share the benefits with as many as we can in a skilful manner. One example (as we make excess capital from trainings, books, consulting, etc) that we are doing soon is nutrient testing eggs and poultry against other products on the market. Whilst this obviously benefits us from a marketing perspective (if the results are good!), it’s not exactly cheap that many could afford it, but I see many will benefit from this when we publish that. We feel it’s the responsibility of any ‘demonstration’ site to actually demonstrate, hence we put so much care and effort into sharing openly and transparently.

I am currently planning on changing my business considerably next year, bringing in a few of the key people who have been with us multiple years (and demonstrated the drive and capacity to manage such systems) as partners in the business, and strive to grow the farm without increasing the size. Stacking out to the max. I think we’re at 50% production in most enterprises currently, and yet already this season we’re already well over the figures presented in older videos. For example, I see we could double Layers, double broilers and double the No-Dig Mkt Gardens production with no increase in size. I’m now curious how far we can push ecosystem processes in a healthy way. Scaling up is very feasible with few people if we shave off some of the distractions that we do for the benefit of the learning context of the farm. For example, I think I will invest in a new sheep flock and lose the cows to make this a more useful enterprise. I might lose the pigs altogether to allow the other half of the farm we don’t really use to regenerate into silvopasture. Other than that we’d be rotating through Broilers/ Hens/ Sheep and then spending most of the day in the Mkt Gardens together, sharing the bookkeeping/accounting/deliveries. We’d all have 3-4 months off (paid) and those 6 winter months would be 2-3 hrs work for whoever’s ‘on’. For context, 10,000 broilers in 6 months gives 4 Swedish salaries, the enterprise investments were all paid off in yr 1. Our Mkt Gardens are planned at 36Eur/m2 net, but I think this year with the change of plan and focus on restaurants we’re kicking 50. I think it could be running at 70 with a couple of caterpillar tunnels for growing on transplants. We could already sell double the eggs tomorrow. People need to see it working, they need to see the figures and need to feel the possibilities and really understand the back-end that makes it all work. The numbers here are really good by any standards, and so I think it’s models worth pushing further and continuing to share openly. ( I’ll be documenting this on our Youtube channel throughout the rest of the season, including summarising the season, finances, etc) Everything we deal with here is all process based, context based, it takes months of carefully targeted support (in my experience) to lay the foundations to really start people off on their entrepreneurial journeys. It’s also why we run such long-term trainings (both internships and our Core-Team). We don’t pay our Core-Team as they are on a long term intensive education program of a nature you can’t just go out and find easily. A LOT of thought and care goes into all aspects what we are doing here, people side included. Our interns pay a chunk, but in context of the timeframe, hours with me and whats on offer here, you realise its very good value when you see it’s cheaper than the cost of backpacking here. We’re having an awesome time, and with this longer time, learning can go very deep from design/ business planning to deep personal development. Short trainings cannot address that stuff effectively, if even at all. Short trainings are much more profitable and require little in the way of responsibility, engagement, people skills, etc, and hence most trainings are short term. We are specifically looking to support more people into farming full time, and hence our choices. The online training we are developing has a very long waiting list already, with feedback that people are really wanting longer term process based learning and mentorship. Looking around there’s not much out there and yet there’s huge demand. I highly recommend Joel Salatins Fields Of Farmers as a book for anyone considering running projects or farms and working with people.

To be honest, I actually want to spend my days farming. I come to all this a bit the opposite way to many. I ran a much more lucrative business designing/ teaching for many years, as I had no costs/ responsibilities. I actually like to graft hard and work with ecosystem processes in a system I craft and steer, so for me I’m choosing to farm for my living and meet my educational objectives by supporting others to get going too. We do make money from books/ trainings/ consults, etc, but it’s like 30%. I’m dedicated to farming smart and taking it further because I’m passionate and energised for that. No-one farms to make excess money, let’s face it; we do this because we’re super passionate about the lifestyle we design for ourselves, the learning, the joyfulness and meaning in all we do.


It is rather a wild myth to think that any labour is free. When you have 30 people using tools/ equipment things degenerate 50 times faster than you working alone with your own gear. What I’d expect from a Swedish wage employee is well above the bar of most. To be honest I’ve had half a dozen folks come through that I’d consider employing. I don’t say that to be harsh, I’m being realistic. I can’t help feeling like I’ve grown up in a generation where a large % of folks lack tangible manual skills, physical strength and endurance, mental clarity, drive and commitment, embodied responsibility and the ability to communicate skilfully. Running a Market Garden is one thing, and complicated enough for many. In our setting, the skill base/ knowledge needed to be a useful employee is quite a lot wider, but then we’re not exactly a ‘normal’ farm. In a place like this, with many hands around, costs really adds up, and it would probably shock folks who don’t do this themselves. We only serve the very finest grass fed meats, pastured that and fully organic/ beyond organic/ biodynamic food. We eat better here than in any place I’ve visited, and living in Sweden that stuff ain’t cheap. This summer I will spend 2000 Eur on butter alone in 5 months! I think it’s too easy to say ‘free labour’ and I’m aware most people who say that do not run farm production businesses or have much experience hosting large groups for extended time periods over many years. Just saying. Many things are faster with people, like mulching all the trees, but actually, most things are very much slower when you have too many. I service egg mobiles alone in 21 mins. During the internship it takes 5 people 45 mins in the beginning. I’m just pointing out that many hands make many responsibilities, and its way more complex than that….

What makes farm-scale permaculture different to permaculture (in my mind) is that we suddenly have economy and regulation to deal with. That’s why some of the idealism has to be dropped in favour of pragmatism. We’re competing in a globalised market place with cheap oil, out of site slave labour, and local agribiz often running on free ag school student labour. If someone is not farming for a living then they’re not really in a clear position to speak to how farms should be run, are they? Luckily, people speak and people lacking integrity ’cull’ themselves, i.e., abusive characters/ places get flagged and highlighted to some degree, and within the professional networks I move within I think a lot of people know whats happening, who/where to recommend or not, etc.

We really do need places like this, because the learning happening here is not stuff you can get in ag school (speaking from my own direct/ other ag school students that come here experience). You can’t get in in short courses, so where to go? You could get it as an employee somewhere like this, where we take a lot of time in learning, but for every such job there must be thousands wanting this learning and experience. If we want a new generation of innovative farmers getting going soon, there’s really no time to mess about. It’s obviously all down to context, which looks different for all of us. We must use our strengths, gifts and talents to step up and contribute in whatever way we can be of benefit, if we decide we want to play a part in the picture.

On another note, reading your blog I saw one of your readers comments, and just want to be clear that I make no financial gain whatsoever from the film. If you do your research you see the funds pay for Olivier and his colleagues work and physical goods. I believe it’s true to say that in his other films there has been some kind of financial arrangements with the other party, but I wanted to openly share another side of the farm to people and do not need to be paid for that.

I guess I waffled on quite a while now. It’s a big topic, and a very important one. Thanks to detailed record keeping here I have much to share on these questions, which is a big part of what we do with the interns here. But in short, our farm is moving towards shared enterprise management/ responsibility/ profit (bringing in equal decision makers as opposed to me having employees). It’s a big step/risk on my part, but I see the opportunity created for incredible folks I love will be life-changing and powerful, and help to take this all to another level in terms of what is possible in small-scale regenerative ag.

Thank you for your engagement and passion to understand these complex yet vital contextual points. I very much look forward to continuing to meet more and more fine folks from around the world who are getting powered up and starting out on this fruitful and rewarding pathway…"



My reply to this was:


"Hi Richard,
Just came back and read your comments. Thank you for replying and being so open about what you do.
I hadn’t got round to going back on my own blog but I was going to say to Dawn that the money was to fund the film making and all that enta
ils. And like I said I was very impressed with the film and don’t regret my small investment in getting it made.
I feel you’ve answered my question very well, it was never about not having somewhere like Ridgedale that was a training farm as well as a working one, it was more asking if it was still viable without the interns. I 100% think that places like yours need to exist, and you’ve justified what you do and the whys of pay very well, do I agree with the pay issue for long term interns? No, but I can see why you do it and you are investing back into areas that would be a benefit to many rather than the few. And I commend you if you are looking at partnerships and profit shares in the future.
I really liked how the film didn’t skirt around this issue and it’s a subject you’ve obviously given a lot of thought and consideration about.
My use of the term “free labour “ was rather lazy, and as a carpenter who’d rather work alone than have the customer “help” me I understand that free labour isn’t always the case. I’m also not in any way criticising the way you keep your interns, to be honest if I was younger (or had less children) it’s certainly something I would consider, your food always looks amazing and they all seem to be very happy and like a family unit. I also place a high value on any learning that I can gain so I can see why you are getting over subscribed.
I also have to confess to being a little obsessed with your videos on YouTube and have to say there’s not many I haven’t watched and I really appreciate them, between you and Curtis Stone that’s all I have time to watch at the moment.
I grew up on a conventional farm where the words “get a real job instead” were used to me, and any farm less than 200 acres wasn’t seen as profitable. But I’ve still maintained my passion for wanting to be a small food producer in a community that wants it, your farm is showing it’s possible, slowly I’m taking steps to get there and I’ve even signed myself up for mentoring at a CSA near to me so I can begin to set up my own very small scale one here.
Could I use your comments here in my next blog post (or to update that one) to show your answers to the questions? I think that’s much better than leaving it open ended.
Thanks again for taking the time to reply, I can see how busy you are!"





          Watch NVIDIA’s CEO Keynote at Computex 2017 here   
NVIDIA AI FORUM: GPU deep learning is bringing on a new era of AI computing. By drawing inspiration from the processes in the human brain, we are making machines that can perceive, understand, and react to the real world. From UAVs to the data center, this technology is being used to help in our lives, […]
          機械学習/Deep Learningが気になる人も要注目、「アルゴリズム」の基本が学べる無料の電子書籍150ページ   

人気過去連載を電子書籍化して無料ダウンロード提供する@IT eBookシリーズ。第29弾では「コーディングに役立つ!アルゴリズムの基本」10回分を1冊のPDFとしてまとめた。アルゴリズムとは何か? なぜ学ぶべきなのだろうか?


          A simple neural network with Python and Keras – PyImageSearch   
Learn how to create a simple neural network using the Keras neural network and deep learning library along with the Python programming language.
          The Future of Radiology and Artificial Intelligence   
There is a lot of hype and plenty of fear around artificial intelligence and its impact on the future of healthcare There are many signs pointing towards the fact that AI will completely move the world of medicine As deep learning algorithms and narrow AI started to buzz especially around the field of medical imaging many radiologists went into panic mode In his presentation at the GPU Tech Conference in San Jose in May 2017 Curtis Langlotz Professor of Radiology and Biomedical Informatics at Stanford University mentioned how he received an e-mail from one of his students saying he was thinking about going into radiology but does not know whether it is a viable profession anymore But the assumption that the radiologist profession is dying is just plain wrong
          Lead Software Engineer - Tenstorrent - Toronto, ON   
Our vision is that the next quantum leap in efficacy of deep learning will be unlocked by joint optimization of both algorithms and the hardware on which they...
From Tenstorrent - Mon, 24 Apr 2017 19:28:13 GMT - View all Toronto, ON jobs
          Processor Architect/Designer - Tenstorrent - Toronto, ON   
Our vision is that the next quantum leap in efficacy of deep learning will be unlocked by joint optimization of both algorithms and the hardware on which they...
From Tenstorrent - Mon, 24 Apr 2017 19:27:36 GMT - View all Toronto, ON jobs
          Deep Learning Expert - Tenstorrent - Toronto, ON   
Our vision is that the next quantum leap in efficacy of deep learning will be unlocked by joint optimization of both algorithms and the hardware on which they...
From Tenstorrent - Fri, 31 Mar 2017 18:03:58 GMT - View all Toronto, ON jobs
          NVIDIA、今年中に 10 万人のディープラーニング開発者をトレーニング予定   
2017 年 5 月 9 日 - NVIDIA は本、AI 分野の専門知識に対する需要の拡大に応えるため、NVIDIA Deep Learning Institute を通じて今年中に 10 万人 (2016 年の 10 倍) の開発者をトレーニングする計画があることを発表しました。
          Comment on White Shark Media: In The Deep Sea Of Success by Lexie Ray   
In the digital marketing world I think everyday is a deep learning environment where the marketers do a lot of research. I would use the service of <a href="https://topaussiewriters.com/aussiessay-com-review/" rel="nofollow">aussie essay review</a> as a way to keep pace with the competitors but learning from other strategies is best. If only I can spot the thing that the big company are looking for then I can make the strike early and have the advantage.
          (Associate) Data Scientist for Deep Learning Center of Excellence - SAP - Sankt Leon-Rot   
Build Machine Learning models to solve real problems working with real data. Software-Design and Development....
Gefunden bei SAP - Fri, 23 Jun 2017 08:50:58 GMT - Zeige alle Sankt Leon-Rot Jobs
          Automated analysis of multi-modal medical data using deep belief networks (DP140102794)   
Dr Gustavo  Carniero has won a 3 year ARC Discovery Grant valued at $295,000. The project will develop an improved breast cancer computer-aided diagnosis (CAD) system that incorporates mammography, ultrasound and magnetic resonance imaging. This system will be based on recently developed deep learning techniques, which have the capacity to process multi-modal data in a […]
          Deep Learning Scientist/Senior Scientist - EXL - Jersey City, NJ   
EXL serves the insurance, healthcare, banking and financial services, utilities, travel, transportation and logistics industries....
From EXL - Tue, 18 Apr 2017 00:09:55 GMT - View all Jersey City, NJ jobs
          Will Nokia Bounce Back with its Cloud Routers?   

Nokia, the embattled cell phone manufacturer, has come up with a new set of routers that it claims will lay the foundation for future technologies such as 5G and the Internet of Things.

In an announcement made by the company, it said that Nokia has made ground-breaking innovations that’ll give more opportunities for service providers to build powerful networks. These are powered by the new FP4 silicon, which is the world’s first multi terabit chipset. It is touted that this chipset is six times more powerful than existing network processors, and this could propel faster speeds for networks.

Two products built on these chipsets are 7750 Service Router and 7950 Extensible Routing System. The first one is expected to transmit 144 terabits per second while the latter can transfer 576 terabits in a second, provided they are in a single system.

Nokia said that these new routing platforms make modern networks more adaptable, faster and safer than before, and at the same time, delivers improved network security and intelligence. These platforms can signal the arrival of the next generation of technologies that include deep learning, machine learning, AI, IoT and more.

A common aspect of all these emerging technologies is the need for handling large amounts of data and “teaching” machines to learn from this data. This obviously requires bigger and faster networks and this is exactly where Nokia is placing its bets.

This brings up the big question of why Nokia chose these routers instead of other lines of business?

For starters, this is one area that Nokia that considerable experience and the other is that it forecasts IP traffic to go to 330 exabytes a month by 2022.

This is why with this new platform, Nokia wants to get back into the business and resurrect its brand name and operations. Though Nokia was one of the leading manufacturers of cell phones, it fell out of favor from customers after the emergence of companies like Samsung that were able to get more powerful smartphones to the market within a short time.

So, is this the right product for Nokia? Will it help the company to become an important player in the technology sector in the future?

Apparently not, according to analysts at Wells Fargo who believe that Nokia needs to do a lot more to get back into the networking business. These analysts contend that service providers are not ready to embrace this high-speed chipsets and platforms yet, so it may take time for Nokia to see the implementation of its new products.

Also, these analysts are skeptical about the ability of service providers to put these platforms to the best use. This means, though Nokia’s new platform can transfer a minimum of 144 terabits per second, will the surrounding infrastructure allow it is a big question.

In the meantime, it makes sense for Nokia to focus on more near-term products that can get some money flowing into its coffers while keeping these platforms as its long-term strategy.

The post Will Nokia Bounce Back with its Cloud Routers? appeared first on Cloud News Daily.


          Nvidia Finds its Niche in Non-gaming Technology   

Nvidia has been a gaming company for a long time, and it has always tied its revenues and business to its gaming hardware.

But, that’s now changing as the company is seeing profits in its artificial intelligence (AI) segment as well. Over the last year and a half, Nvidia realized that it can go beyond its traditional gaming business.

These efforts are evident in the first quarter results of its 2018 fiscal year. In fact, its traditional gaming business performed less than expected. It earned a revenue of only $1.03 billion against the average forecast of $1.13 billion.

During this same time, its data-center business saw a big boost in revenue. It reported an earnings of $400 million, which is close to what the company earned in the entire fiscal year of 2016. This goes to show the growth of its data center business over the last one and half years. Besides its data center, its self-driving and automation division also saw a big jump in revenue.

A deep analysis reveals some interesting trends for the company. Firstly, it’s moving away from its core business slowly and steadily, as the loss in its gaming division was made up by the buoyant revenue from its AI and data center divisions. In fact, this expansion into other areas was given a big thumbs-up by the investors. As soon as the results were announced, the stock price of Nvidia went for a joy ride.

Secondly, the company’s move came at a right time when cloud computing companies are vying with each other to woo customers. In the process, they want to offer products with faster processing speeds. This requires chipsets with advanced deep learning and AI technology, something that Nvidia was able to cater to.

A press release by the company said that it attributed much of its efforts in cloud due to the adoption of its chipsets by some of the largest companies in the world such as Amazon Web Services (AWS), Alphabet Inc, Microsoft, Facebook, IBM and Alibaba Group Holding.

If you’re wondering what’s special about Nvidia’s chipsets, well nothing actually.

The Graphics Processing Units (GPUs) were initially being used for 3D rendering and for gaming. Soon, cloud companies realized that the same chip can be used for other processes too as they have high processing power. So, they were adopted by these companies to increase their computing power and that’s how Nvidia’s GPUs became a much sought after product.

Going forward, almost every major cloud provider is looking to standardize the use of GPUs, and this is definitely good news for Nvidia. For its investors and management, it means another few years of bounty results and less dependence on the changing gaming industry. One of the drawbacks of the gaming industry is that it is cyclical, with sales soaring  through the holiday season, but remaining subdued through the rest of the year. This move to the cloud means the company can no longer worry about it.

Once again, these results and trends show the over-reaching impact of cloud technology across [...]

The post Nvidia Finds its Niche in Non-gaming Technology appeared first on Cloud News Daily.


          Vincent Granville posted a blog post   
Vincent Granville posted a blog post

          Il deep learning per la protezione delle coltivazioni   

L’utilizzo delle biotecnologie in agricoltura è un tema spinoso e in questo senso non sono mancate e non mancano le critiche alle grandi corporation del settore.

Una nuova iniziativa di Monsanto potrebbe però portare vantaggi a tutti gli agricoltori: applicare l’intelligenza artificiale per scoprire velocemente nuove sostanze che proteggano le coltivazioni da parassiti e malattie.

Il progetto si concretizza attraverso una collaborazione con Atomwise, società che già usa il deep learning per varie applicazioni in campo biochimico.

Il punto di partenza del progetto è la constatazione che i normali processi per la creazione di nuove sostanze in agricoltura sono molto lenti.

Mediamente lo sviluppo di un nuovo prodotto per la protezione delle coltivazioni richiede undici anni e 250 milioni di dollari di investimenti prima di arrivare alla commercializzazione.

Il deep learning di AtomNet "testa" un possibile ligando (in viola)

L’idea è quella di applicare algoritmi di deep learning alla scoperta di nuove molecole e delle loro interazioni. A questo serve la tecnologia di Atomwise, che ha realizzato un algoritmo specifico per la chimica molecolare, battezzato AtomNet e basato su reti neurali.

AtomNet nel tempo ha “imparato” le varie possibili interazioni fra molecole e in particolare quando una certa molecola può essere il giusto ligando per una determinata proteina-bersaglio, ossia quando può legarsi con quest’ultima e generare un effetto biologico.

Avendo “assorbito” questo principio generale della biochimica a partire dalle scoperte già effettuate, AtomNet può assere di grande aiuto in particolare nella fase iniziale di studio di una nuova molecola.

Il suo ruolo sarà infatti prevedere quali molecole potrebbero avere un effetto positivo nella prevenzione di parassiti e malattie per le coltivazioni, consentendo di scartare a priori quelle che non appaiono adatte. Invece di effettuare questa selezione con prove dirette, è l’algoritmo a dare la direzione in cui conviene indirizzare la ricerca.

L'articolo Il deep learning per la protezione delle coltivazioni è un contenuto originale di 01net.


          The Rise of Artificial Intelligence in Events [Webinar]   

Join us for this free webinar to learn how to get more from artificial intelligence for your event. Chatbots, Deep Learning, Concierge, Big Data. What all these words hide is an incredible opportunity for event professionals willing to embrace innovation. Artificial Intelligence offers revolutionary opportunities to grow your event. Whether it is customer support, event management, marketing, […]

The post The Rise of Artificial Intelligence in Events [Webinar] by Julius Solaris appeared first on http://www.EventManagerBlog.com


          O’Reilly AI Conference News Roundup   

O'Reilly AI Conference news roundup: TensorFlow support, deep learning, and AI training data were some notable announcements.

The post O’Reilly AI Conference News Roundup appeared first on RTInsights.


          Slides: Machine Learning Summer School @ Max Planck Institute for Intelligent Systems, Tübingen, Germany   


Here are the slides of some of the presentations at the Machine Learning Summer School at the Max Planck Institute for Intelligent Systems, Tübingen, Germany

Shai Ben-David
(Waterloo)
 Learning Theory.
Slides part 1 part 2 part 3
Dominik Janzing
(MPI for Intelligent Systems)
 Causality.
Slides here.
Stefanie Jegelka
(MIT)
 Submodularity.
Slides here.


Jure Lescovec
(Stanford)
Network Analysis.
Slides 1 2 3 4




Ruslan Salakhutdinov
(CMU)
Deep Learning.
Slides part 1 part 2


Suvrit Sra
(MIT)
Optimization.
Slides 1 2 3A 3B
Bharath Sriperumbudur
(PennState)
Kernel Methods.
Slides part 1 part 2 part 3


Max Welling
(Amsterdam)
Large Scale Bayesian Inference with an Application to Bayesian Deep Learning
Slides here.




Bernhard Schölkopf
(MPI for Intelligent Systems)
Introduction to ML and speak on Causality.
Slides here.

h/t Russ




Join the CompressiveSensing subreddit or the Google+ Community or the Facebook page and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

          Pixlab   
Process and analyze input media images or video content using the PixLab Rest API. It uses built-in HTTP capabilities for passing parameters and authentication that responds with standard HTTP response codes. It allows you to process, transform and filter any images from any programming language with machine vision and deep learning APIs. The API returns JSON by default or BLOB Image Content on demand, with results for Image Processing, Machine Vision and Media Analysis. This includes; facedetect, tagimg, facelookup, encrypt, decrypt and more. PixLab provides media storage with scalable and unified RESTful APIs for media analysis and processing tasks.
Date Updated: 2017-06-29
Tags: Media, Analytics, , Applications, , Big Data, , Content, , Identity, , Images, , Machine Learning, , Meme, , Mobile, , Recognition, , Storage

          Deep Learning Library Software Development Engineer - NVIDIA - Polska   
You will be interacting with members of the open source deep learning software engineering community to define and implement the new features of the CUDNN...
Od NVIDIA - Thu, 22 Jun 2017 07:34:43 GMT - Pokaż wszystkie Polska oferty pracy
          Deep Learning Software Development Engineer - NVIDIA - Polska   
Work directly with deep learning framework developers by collaborating on open source code bases. Ability to work independently, define project goals and scope,...
Od NVIDIA - Thu, 22 Jun 2017 07:34:27 GMT - Pokaż wszystkie Polska oferty pracy
          Building Conversation Bots with Amazon Lex   
         

Using voice and text you can now integrate Amazon Lex in your chat bots using AI ( artificial intelligence & deep learning ). This allows for our client to have better customer support with life like natural conversations that answer real problems. So for instance your looking to book a hotel you ask I would like to book a hotel, Lex answers sure where would you like to book, you answer New York, Lex answers when is your stay and keeps answering your questions and collecting the data until the fulfillment mechanism for you intent is completed. By Lex fulfilling the intent of users questions it can prompt to then verify the user input. The future of chat bots is bright as two thirds of consumers now interact with companies via chat so the natural leveraging of using chat bots  for repetitive tasks such as looking up shipping information, warranty information or other repeated tasks within the organization. With AI technologies emerging all the time Amazon Lex is nice addition to voice and speech  allowing for common uses as chat bots, application bots, transactional bots, productivity bots and device controllers.

         
          (USA-NY-Brooklyn) Projoect Edision Project Manager   
JPMorgan Chase & Co. (NYSE: JPM) is a leading global financial services firm with assets of $2.5 trillion and operations in more than 60 countries. The firm is a leader in investment banking, financial services for consumers, small business and commercial banking, financial transaction processing, asset management, and private equity. Senior, hands-on development manager who is passionate about technology and has experience developing high performance transaction or reporting systems with large databases and multi-tier architectures; knows how to build and motivate a talented and committed technology team The firm is on a journey to transform the way we deliver technology, and to drive business value from our data. As such, the Corporate and Investment Bank (CIB) is looking for an experienced Project Manager for our Edison program. Edison is a scalable decision support system that utilizes Big Data technology, micro-service oriented architecture, and “inner sourcing” development model. It will store master reference data, support reporting, BI, deep learning, and data science and provide data governance. + Develop project plans and execution approaches for a number of initiatives that are defined by client requirements and specifications. + Work with other cross-functional managers to select a core team for each project, present a plan and gain approval from the business and IT leadership. + Monitor ongoing projects to evaluate progress and take action if any issues arise. + Work with QA Manager and testing team to ensure thorough testing of modules/applications/interfaces. + See projects to completion within terms of SLA and supervise accurate turnover to Operations staff. + Oversee performance management of assigned team, giving feedback, writing appraisals and approving development plans with senior team members. **What Education/Experience do I Need?** + You have at least 5 years of combined business, project management, team leadership and IT experience. + You should have a college degree or equivalent specialized training also. + Must possess experience managing geographically distributed and culturally diverse workgroups and demonstrate solid team management, leadership and coaching skills. + Excellent communication, analytical and writing skills; can handle immense amounts of work and tight deadlines; able to develop strong client relationships; confident when working with professional services firms. **Other Requirements** CMM and Six Sigma methodologies and standards required along with Microsoft Project, Project Workbench and PMI required. **Competitive Rewards** Competitive pay; excellent benefits; a performance-based incentive program; opportunity for company advancement. **Work Environment** Dynamic, no-nonsense business center. + College degree, specialized training or equivalent work experience + Minimum five years of combined business, project management, team leadership and IT experience required + Experience with projects in multiple technologies, functions (e.g. transaction management, risk management etc.) and industries + Knowledge of CMM and Six Sigma Methodologies and Standards + Knowledge and experience using project management software such as Microsoft Project, Project Workbench, PMI, etc. + Experience managing geographically distributed and culturally diverse work-groups with strong team management, leadership and coaching skills + Project Management Certification a plus + Knowledge of outsourcing methodologies and operating models, and working with professional services firms + Excellent written and verbal communication skills + Ability to develop strong client relationships JPMorgan Chase offers an exceptional benefits program and a highly competitive compensation package. JPMorgan Chase is an Equal Opportunity and Affirmative Action Employer, M/F/D/V JPMorgan Chase is an equal opportunity and affirmative action employer Disability/Veteran.
          Deep Learning and GPU Acceleration in Hadoop 3.0   

Recently Raj Verma (President & COO of Hortonworks) spoke to Jim McHugh from Nvidia at the DataWorks Summit keynote in San Jose (video). Jim began by talking about how parallel processing that is used in gaming is also essential to Deep Learning*. And the lifeblood of Deep Learning is data. With its insatiable desire for […]

The post Deep Learning and GPU Acceleration in Hadoop 3.0 appeared first on Hortonworks.


          Lead Software Engineer - Tenstorrent - Toronto, ON   
Our vision is that the next quantum leap in efficacy of deep learning will be unlocked by joint optimization of both algorithms and the hardware on which they...
From Tenstorrent - Mon, 24 Apr 2017 19:28:13 GMT - View all Toronto, ON jobs
          Processor Architect/Designer - Tenstorrent - Toronto, ON   
Our vision is that the next quantum leap in efficacy of deep learning will be unlocked by joint optimization of both algorithms and the hardware on which they...
From Tenstorrent - Mon, 24 Apr 2017 19:27:36 GMT - View all Toronto, ON jobs
          Deep Learning Expert - Tenstorrent - Toronto, ON   
Our vision is that the next quantum leap in efficacy of deep learning will be unlocked by joint optimization of both algorithms and the hardware on which they...
From Tenstorrent - Fri, 31 Mar 2017 18:03:58 GMT - View all Toronto, ON jobs
          智慧云中的FPGA    
作者: hyukhae079408 发表于: 2017-06-28 14:30:03 (GMT 8) 简介: [color=#31424e][backcolor=rgb(247, 247, 247)][font=宋体][size=2]人工智能大热之前,Cloud或Data Center已经开始使用FPGA做各种加速了。而随着Deep Learning的爆发,这种需求越来越强劲。本文主要讨论Cloud巨头Amazon和Microsoft的FPGA策略。[/size][/font][/backcolor
          #heiseshowXXL: Revolution der KI mit Deep Learning   
Aufzeichnung der #heiseshowXXL von der CeBIT 2017 zum Thema „Revolution der KI mit Deep Learning“ mit Oliver Zeigermann.
          Deep Learning on-Chip   

Summer school
20-22 September 2017 - Politecnico di Torino, corso Duca degli Abruzzi, 24 - Torino


           Radeon Instinct MI25, MI8 και MI6 ανακοίνωσε η AMD   
Σε πρόσφατη εκδήλωση που πραγματοποίησε η AMD, η οποία ήταν αφιερωμένη στη νέα επιχειρησιακής κλάσης σειρά επεξεργαστών EPYC 7000 Series, η AMD παρουσίασε και τρεις deep learning accelerators. Οι τρεις νέοι επιταχυντές deep learning δεν απευθύνονται βεβαίως σε καταναλωτές, με τον πυρήνα της κορυφαίας Radeon Instinct MI25 που βασίζεται στην αρχιτεκτονική Vega να ενσωματώνει 64 μονάδες compute “επόμενης γενιάς” που είναι ικανές για 24.6 TFLOPS απόδοσης Half Precision Floating Point (12.3 TFLOPS Μονής Ακρίβειας). 
Οι συγκεκριμένες κάρτες γραφικών προορίζονται για εφαρμογές deep learning και machine AI και η κορυφαία πρόταση της AMD δεν έρχεται με λιγότερα από 16GB μνήμης HBM2, για memory bandwidth ύψους 484GB/s. 
Οι Radeon Instinct MI8 και MI6 βασίζονται στους πυρήνες Fiji και Polaris με την πρώτη να διαθέτει 64 μονάδες compute (4096 stream processors) και 4GB μνήμης HBM1 για 512GB/s memory bandwidth και 8,2 TFLOPS είτε FP16 είτε FP32, Half ή Single Precision Compute. H Radeon Instinct MI6 από την άλλη που βασίζεται στην αρχιτεκτονική GPU με την ονομασία Polaris διαθέτει 36 μονάδες computing και 2304 stream processors αλλά και 16GB μνήμης GDDR5 για memory bandwidth στα 224GB/s και απόδοση Half ή Single Precision, FP16 ή FP32 στα 5,7TFLOPs. insomnia
          Data Scientist (Artificial Intelligence & Deep Learning) - Microsoft - Redmond, WA   
We help drive actionable business intelligence through advanced statistical modeling and business analytics across Microsoft....
From Microsoft - Sat, 13 May 2017 05:34:57 GMT - View all Redmond, WA jobs
          AI Podcast: How Deep Learning Can Translate American Sign Language   

Deep learning has accelerated machine translation between spoken and written languages. But it’s lagged behind when it comes to sign language. Now Syed Ahmed, a computer engineering major at the Rochester Institute of Technology, is unleashing its power to translate between sign language and English. “You want to talk to your deaf or hard of […]

The post AI Podcast: How Deep Learning Can Translate American Sign Language appeared first on The Official NVIDIA Blog.


          How a Boston Startup Plans to Poach Poachers in Africa Using Intelligent Drones   

Across the African savanna, 25,000 elephants and 1,000 rhinos are killed by poachers each year, according to some estimates. At this rate, they’ll be extinct within two decades. To combat this crisis, Neurala, a Boston-based startup, is bringing stealth, speed and scale to the fight in the form of deep learning-powered drones. By putting intelligent […]

The post How a Boston Startup Plans to Poach Poachers in Africa Using Intelligent Drones appeared first on The Official NVIDIA Blog.


          Amazon Lex now supports 8 kHz telephony audio for increased speech recognition accuracy   

The Amazon Lex speech recognition engine has now been trained on telephony audio (8 kHz) to provide increased speech recognition accuracy for conversations over the phone. Amazon Lex enables your users to interact with your application via natural conversation using the same deep learning technology as Amazon Alexa to fulfill most common requests. Amazon Lex chatbots maintain context and manage the dialogue, dynamically adjusting the responses based on the conversation. 


          Inteligência artificial pôs à prova psicografia de Chico Xavier   

chico-xavier-wikimedia-commons

Publicado na Exame

Francisco Cândido Xavier morreu há 15 anos, deixando para trás mais de 412 livros escritos. Mas ele sempre rejeitou a autoria de todos: a obra seria inteira psicografada, ditada diretamente de espíritos que falavam ao médium.

Com o aniversário de falecimento do líder espírita, uma empresa brasileira resolveu investigar a obra de Chico usando inteligência artificial. Ao longo da vida, ele psicografou livros de vários autores diferentes. A ideia era usar todo o poder de computação para responder duas perguntas: esse autores têm cada um seu estilo próprio? Eles são suficientemente diferentes entre si?

A Stilingue, uma empresa que trabalha com análise de textos via inteligência artificial para “resumir a internet”, encontrando tendências nas redes sociais, resolveu testar como as obras psicografadas seriam analisadas por uma técnica de aprendizado de máquinas chamada Deep Learning.

A partir de grandes quantidades de dados, o computador aprende a criar relações entre eles, sem precisar aprender, por exemplo, o que é um verbo, um adjetivo, um substantivo. Se fosse reconstruir a Bíblia, o computador logo ia aprender que precisa colocar um número antes de cada frase, porque o livro é estruturado em versículos.

A mesma técnica também já foi usada para recriar Shakespeare. Depois de ler milhões de caracteres do dramaturgo, o computador era capaz de escrever sozinho “imitando” o estilo do inglês, sem nunca ter passado por uma aula de literatura. Nem sempre as frases fazem total sentido, mas os tempos verbais e a mania de criar palavras novas mudando o final delas ficam reproduzidos, igualzinho.

No caso de Chico Xavier, o estudo da Stilingue selecionou três dos principais autores psicografados pelo médium: Emmanuel, André Luiz e Humberto de Campos.

Para “alimentar” a rede neural artificial, eles selecionaram três livros de cada autor – que precisam ser enormes, porque a técnica deep learning exige, no mínimo, um milhão de caracteres por autor conseguir aprender com sucesso. “No caso de Humberto de Campos, sentimos um pouco de falta de mais material. Ele é um autor mais desafiador porque escrevia diferentes tipos de texto [contos, anedotas e poesias]”, explica Milton Stiilpen Jr., fundador da Stilingue.

Devidamente treinado, o computador começou a reproduzir os textos. André Luiz, por exemplo, tinha o hábito de colocar falas espaçadas entre blocos de texto maiores, ao invés de criar longos blocos de diálogos.

André Luiz: entidade espírita vs. bot

Este primeiro texto foi psicografado por Chico Xavier

Os encarnados presentes viam tão-somente o corpo de Otávia, dominado pelo sacerdote que lhes era invisível, quase a rebentar-se de soluços atrozes, mas nós víıamos além. A nobre senhora desencarnada postou-se ao lado do filho e começou a beijá-lo, em lágrimas de reconhecimento e amor. Pranto copioso identificava-os. Cobrando forças novas, a genitora continuou:

– Perdoe-me, filho querido, se noutra época induzi o seu coração à responsabilidade eclesiástica, modificando o curso de suas tendências. Suas lutas de agora me atingem a alma angustiada. Seja forte, Marinho, e ajude-me! Desvencilhe-se dos maus companheiros! Não vale rebelar-se. Nunca fugiremos à lei do Eterno! Onde você estiver, a voz divina se fará ouvir no imo da consciência…

Nesse momento, observei que o sacerdote recordou instintivamente os amigos, tocado de profundo receio. Agora que reencontrava a mãezinha carinhosa e devotada a Deus, que sentia a vibração confortadora do ambiente de fraternidade e féé, sentia medo de regressar ao convívio dos colegas endurecidos no mal.

Já este foi criação da inteligência artificial

A primeira vez mais providencial de serviço de sua consciência, a senhora Laura encontrava-se com a presença de alguns, com a sua consciência espiritual e a medicina de amor, acrescentou:

– O controlador de serviço está disposto a escapar com as mesmas expressões de alegria.

A primeira vez mais forte de algum tempo, a senhora de Alexandre prosseguiu a companheira de serviço e considerando a alegria da conversação despediu-se:

– Neste momento, a maioria dos companheiros encarnados estão através de construções destruidoras e desencarnadas. A consciência tem sempre a construção do coração.

Depois de criar três bots capazes de imitar os autores com uma precisão considerável (erro de 22% para André Luiz, 5% para Emmanuel e 32% para o Humberto de Campos), dá para dizer que cada autor tem um estilo razoavelmente marcante e uniforme.

Agora, dá para dizer que eles são diferentes entre si? Ou será que o estilo delata que teriam sido escritos por uma só pessoa? Para fazer o teste, eles decidiram confundir a máquina. Misturaram os textos de diferentes autores. Mandaram o bot do Emmanuel escrever com base na obra do Humberto, o do Humberto imitar o André e assim por diante. Deu errado: a taxa de erro disparou. Os modelos eram incapazes de encontrar os mesmos padrões de estilo de uma entidade espírita nos livros da outra. Os autores são, sim, marcadamente diferentes.

A questão que resta é: há outras formas de explicar o resultado?

Misturar textos de diferentes temas e épocas de um mesmo autor já é suficiente para aumentar a taxa de erro. Mas não tanto assim. “Fizemos um teste com o Paulo Coelho justamente para testar um único autor com diferentes livros e muitos textos. A taxa de erro aumenta – mas mesmo assim continua baixa”, explica Milton. O teste com Paulo Coelho retornou uma taxa de apenas 10%.

Outra possibilidade cética seria a criação consciente e deliberada de Chico Xavier de diferentes personas, uma para cada autor – coisa parecida com o que o escritor Fernando Pessoa fez, com seis heterônimos marcadamente diferentes.

Milton também tinha uma resposta para isso: eles fizeram o teste de deep learning também com Fernando Pessoa. “Faltou quantidade de dados suficiente para atender essa técnica”, responde Stiilpen. A Stilingue não conseguiu acesso fácil e digitalizado à quantidade necessária de material de cada heterônimo de Pessoa. Relembrando, o mínimo necessário para a análise usando deep learning é de 1 milhão de caracteres o que significa, nesse caso, 6 milhões para uma análise de todos os “autores” em questão. E isso só para aquecer.

Graças a esses resultados, a análise textual deve virar um projeto de pesquisa oficial que vai, inclusive, selecionar outras técnicas mais adequadas a autores como Fernando Pessoa e Nelson Rodrigues. Mas, de tudo isso, qual foi o veredito do estudo sobre Chico Xavier?

A psicografia segue como uma questão de fé. Mas se o estudo atesta algo, é a genialidade do médium. Escrever o volume de texto que ele escreveu, com personas comprovadamente distintas, mas uniformes entre si, não precisa nem ser sobrenatural para ser absolutamente impressionante. Ou, como colocou Monteiro Lobato, “Se Chico Xavier produziu tudo aquilo por conta própria, então ele merece ocupar quantas cadeiras quiser na Academia Brasileira de Letras.”


          Inteligência artificial pôs à prova psicografia de Chico Xavier   

Redes neurais artificiais analisaram obras do médium, morto há exatos 15 anos. E concluíram: fé à parte, Chico era um fenômeno mesmo.

chico-xavier

Ana Carolina Leonardi, na Superinteressante

Francisco Cândido Xavier morreu há 15 anos, deixando para trás mais de 412 livros escritos. Mas ele sempre rejeitou a autoria de todos: a obra seria inteira psicografada, ditada diretamente de espíritos que falavam ao médium.

Com o aniversário de falecimento do líder espírita, uma empresa brasileira resolveu investigar a obra de Chico usando inteligência artificial. Ao longo da vida, ele psicografou livros de vários autores diferentes. A ideia era usar todo o poder de computação para responder duas perguntas: esse autores têm cada um seu estilo próprio? Eles são suficientemente diferentes entre si?

A Stilingue, uma empresa que trabalha com análise de textos via inteligência artificial para “resumir a internet”, encontrando tendências nas redes sociais, resolveu testar como as obras psicografadas seriam analisadas por uma técnica de aprendizado de máquinas chamada Deep Learning.

A partir de grandes quantidades de dados, o computador aprende a criar relações entre eles, sem precisar aprender, por exemplo, o que é um verbo, um adjetivo, um substantivo. Se fosse reconstruir a Bíblia, o computador logo ia aprender que precisa colocar um número antes de cada frase, porque o livro é estruturado em versículos.

A mesma técnica também já foi usada para recriar Shakespeare. Depois de ler milhões de caracteres do dramaturgo, o computador era capaz de escrever sozinho “imitando” o estilo do inglês, sem nunca ter passado por uma aula de literatura. Nem sempre as frases fazem total sentido, mas os tempos verbais e a mania de criar palavras novas mudando o final delas ficam reproduzidos, igualzinho.

No caso de Chico Xavier, o estudo da Stilingue selecionou três dos principais autores psicografados pelo médium: Emmanuel, André Luiz e Humberto de Campos.

Para “alimentar” a rede neural artificial, eles selecionaram três livros de cada autor – que precisam ser enormes, porque a técnica deep learning exige, no mínimo, um milhão de caracteres por autor conseguir aprender com sucesso. “No caso de Humberto de Campos, sentimos um pouco de falta de mais material. Ele é um autor mais desafiador porque escrevia diferentes tipos de texto [contos, anedotas e poesias]”, explica Milton Stiilpen Jr., fundador da Stilingue.

Devidamente treinado, o computador começou a reproduzir os textos. André Luiz, por exemplo, tinha o hábito de colocar falas espaçadas entre blocos de texto maiores, ao invés de criar longos blocos de diálogos.

André Luiz: entidade espírita vs. bot

Este primeiro texto foi psicografado por Chico Xavier

Os encarnados presentes viam tão-somente o corpo de Otávia, dominado pelo sacerdote que lhes era invisível, quase a rebentar-se de soluços atrozes, mas nós víıamos além. A nobre senhora desencarnada postou-se ao lado do filho e começou a beijá-lo, em lágrimas de reconhecimento e amor. Pranto copioso identificava-os. Cobrando forças novas, a genitora continuou:

– Perdoe-me, filho querido, se noutra época induzi o seu coração à responsabilidade eclesiástica, modificando o curso de suas tendências. Suas lutas de agora me atingem a alma angustiada. Seja forte, Marinho, e ajude-me! Desvencilhe-se dos maus companheiros! Não vale rebelar-se. Nunca fugiremos à lei do Eterno! Onde você estiver, a voz divina se fará ouvir no imo da consciência…

Nesse momento, observei que o sacerdote recordou instintivamente os amigos, tocado de profundo receio. Agora que reencontrava a mãezinha carinhosa e devotada a Deus, que sentia a vibração confortadora do ambiente de fraternidade e féé, sentia medo de regressar ao convívio dos colegas endurecidos no mal.

Já este foi criação da inteligência artificial

A primeira vez mais providencial de serviço de sua consciência, a senhora Laura encontrava-se com a presença de alguns, com a sua consciência espiritual e a medicina de amor, acrescentou:

– O controlador de serviço está disposto a escapar com as mesmas expressões de alegria.

A primeira vez mais forte de algum tempo, a senhora de Alexandre prosseguiu a companheira de serviço e considerando a alegria da conversação despediu-se:

– Neste momento, a maioria dos companheiros encarnados estão através de construções destruidoras e desencarnadas. A consciência tem sempre a construção do coração.

 

Depois de criar três bots capazes de imitar os autores com uma precisão considerável (erro de 22% para André Luiz, 5% para Emmanuel e 32% para o Humberto de Campos), dá para dizer que cada autor tem um estilo razoavelmente marcante e uniforme.

Agora, dá para dizer que eles são diferentes entre si? Ou será que o estilo delata que teriam sido escritos por uma só pessoa? Para fazer o teste, eles decidiram confundir a máquina. Misturaram os textos de diferentes autores. Mandaram o bot do Emmanuel escrever com base na obra do Humberto, o do Humberto imitar o André e assim por diante. Deu errado: a taxa de erro disparou. Os modelos eram incapazes de encontrar os mesmos padrões de estilo de uma entidade espírita nos livros da outra. Os autores são, sim, marcadamente diferentes.

A questão que resta é: há outras formas de explicar o resultado?

Misturar textos de diferentes temas e épocas de um mesmo autor já é suficiente para aumentar a taxa de erro. Mas não tanto assim. “Fizemos um teste com o Paulo Coelho justamente para testar um único autor com diferentes livros e muitos textos. A taxa de erro aumenta – mas mesmo assim continua baixa”, explica Milton. O teste com Paulo Coelho retornou uma taxa de apenas 10%.

Outra possibilidade cética seria a criação consciente e deliberada de Chico Xavier de diferentes personas, uma para cada autor – coisa parecida com o que o escritor Fernando Pessoa fez, com seis heterônimos marcadamente diferentes.

Milton também tinha uma resposta para isso: eles fizeram o teste de deep learning também com Fernando Pessoa. “Faltou quantidade de dados suficiente para atender essa técnica”, responde Stiilpen. A Stilingue não conseguiu acesso fácil e digitalizado à quantidade necessária de material de cada heterônimo de Pessoa. Relembrando, o mínimo necessário para a análise usando deep learning é de 1 milhão de caracteres o que significa, nesse caso, 6 milhões para uma análise de todos os “autores” em questão. E isso só para aquecer.

Graças a esses resultados, a análise textual deve virar um projeto de pesquisa oficial que vai, inclusive, selecionar outras técnicas mais adequadas a autores como Fernando Pessoa e Nelson Rodrigues. Mas, de tudo isso, qual foi o veredito do estudo sobre Chico Xavier?

A psicografia segue como uma questão de fé. Mas se o estudo atesta algo, é a genialidade do médium. Escrever o volume de texto que ele escreveu, com personas comprovadamente distintas, mas uniformes entre si, não precisa nem ser sobrenatural para ser absolutamente impressionante. Ou, como colocou Monteiro Lobato, “Se Chico Xavier produziu tudo aquilo por conta própria, então ele merece ocupar quantas cadeiras quiser na Academia Brasileira de Letras.”


          #10: Deep Learning (Adaptive Computation and Machine Learning)   
Deep Learning
Deep Learning (Adaptive Computation and Machine Learning)
Ian Goodfellow , Yoshua Bengio , Aaron Courville
(3)

Neu kaufen: EUR 70,99 EUR 69,99
44 Angebote ab EUR 62,14

(In der Sachbücher-Bestseller-Liste finden Sie maßgebliche Informationen über die aktuelle Rangposition dieses Produkts.)
          Facundo Batista: PyCon Argentina 2016   


El fin de semana pasado fue la octava edición de la conferencia nacional de Python en Argentina. Se realizó en Bahía Blanca, tres días de talleres y charlas.

Yo dí una charla, "Bindings, mutable default arguments, y otros quilom... detalles", y asistí a otras; las que más me gustaron fueron "Poniéndonos un poco más serios con Kivy" por Sofía Martin y alguien más que no recuerdo, "Compartiendo memoria eficientemente con proxies" por Claudio Freire, "Argentina en Python: comunidad, sueños, viajes y aprendizaje" por Humitos, "MicroPython en EDU-CIAA" por Martín Ribelotta, "Redes neuronales con Python utilizando Keras" por Fisa, "Deep learning: aprendiendo con la escafandra" por Javi Mansilla, e "Introducción a programación paralela con PyOpenCL" por Celia Cintas.

Mi charla, renovada

Las keynotes estuvieron muy bien, también. Fernando Schapachnik, de la Fundación Sadosky nos habló del problema de género en las comunidades informáticas (con datos, análisis, y una arenga política al final que estuvo bárbara). Ángel Medinilla nos dío una charla-show-standup sobre metodologías ágiles (excelente presentación). Y la última fue de Victoria Martínez de la Cruz, contando las ventajas y desventajas de trabajar de forma remota (algo que se está imponiendo más y más en las comunidades de software y que está lleno de mitos, así que era muy necesaria).

La organización del evento también estuvo impecable. Se nota que laburaron un montón y salió todo muy bien.

Los asistentes a punto de escuchar una plenaria

Más allá del costado técnico, y de lo que sucede en estos eventos de charlas que se generan, reencuentros, etc, tanto en pasillos como luego de la conferencia en bares o por ahí, quiero destacar el lado "humano"que tuvo esta conferencia.

No sólo las keynotes hablaron de las personas o sus grupos de trabajo, sino que también tuvimos charlas que hicieron lagrimear a varios, como la de Humitos que mencioné arriba o la de Roberto Alsina ("Cómo desarrollar software libre (o no) y no morir en el intento (o no)", que no pude ver pero me contaron). Pero había algo más en el ambiente. Gente comentando lo copada que son organizadores y asistentes en este evento, que cómo te ayudan con todo, que se preocupan, etc. Había muy buena onda por todos lados.

Relajando un poco, en el almuerzo del primer día

Trabajando en uno de los espacios abiertos que había

Hubo una anécdota interesante, también. Resulta que una señora vio en un kiosco a unos asistentes a la conferencia que tenían algo de Python encima. Entonces fue a la escuela de su hijo mayor, de 13 años, lo sacó antes de hora y volvieron a la zona del kiosco (que obviamente, era muy cerca del edificio de la conferencia). Justo pasábamos otros chicos y yo, vieron un pin de Python que llevo en la mochila, y nos preguntaron qué onda. Les contamos de la conferencia, Diego M. les regaló el librito del evento, y listo.

Nosotros pensábamos que terminaba ahí. Nada más lejos.

Al rato volvemos al edificio donde se desarrollaba el evento y vemos que sube a la zona de la conferencia la madre y los dos niños. El pibe de 13 se colgó todo el día yendo de charla en charla, mientras la mamá le hacía el aguante en una zona con sillones. No sólo eso, sino que fueron el sábado y el domingo a la conferencia, y se pasaron todo el finde allí. Notable.

Todas las manos todas

Para cerrar les dejo las fotos que saqué, más esta búsqueda de tuiter que está buena.


          NVIDIA: Startup to Poach Poachers Using Intelligent Drones   
Across the African savanna, 25,000 elephants and 1,000 rhinos are killed by poachers each year, according to some estimates. At this rate, they’ll be extinct within two decades. To combat this crisis, Neurala, a Boston-based startup, is bringing stealth, speed and scale to the fight in the form of deep learning-powered drones. By putting intelligent […]
          Công nghệ thông tin - Hybrid Apps Developer   
CONG TY TNHH GIAI PHAP PHAT TRIEN DOANH SO THONG - Tp Hồ Chí Minh - Native is plus Mô tả Lập trình xây dựng các hệ thống của công ty và ứng dụng di động, hệ thống Big Data + Deep learning Quyền lợi - Môi...
          Công nghệ thông tin - Mobile Apps Developer (Android / Ios)   
CONG TY TNHH GIAI PHAP PHAT TRIEN DOANH SO THONG - Tp Hồ Chí Minh - Native is plus Mô tả Lập trình xây dựng các hệ thống của công ty và ứng dụng di động, hệ thống Big Data + Deep learning Quyền lợi - Môi...
          TensorFlow: resoconto secondo incontro Meetup "Machine-Learning e Data Science" Roma (19 febbraio 2017)   

170219-tensorflow.jpg

Il 16 febbraio 2017 si è svolto a Roma – presso il Talent Garden di Cinecittà - il secondo incontro del Meetup "Machine Learning e Data Science" (web, fb, slideshare): l’incontro - organizzato insieme al Google Developer Group Roma Lazio Abruzzo   - è stato dedicato alla presentazione di TensorFlow,  la soluzione di Google per il deep learning nel machine learning.

Nel seguito una breve sintesi dell’incontro del 16 febbraio 2017.

Prima di iniziare:

•    Cos’è il Machine Learning? Intervista a Simone Scardapane sul Machine Learning Lab (1 dicembre 2016)
•    Cos’è TensorFlow? Andrea Bessi, “TensorFlow CodeLab”, Nov 16, 2016

Resoconto

Premessa

Il 15 Febbraio 2017 a Mountain View (USA) si è tenuto il “TensorFlow Dev Summit” , il primo evento ufficiale di Google dedicato a TensorFlow, la soluzione di deep learning rilasciato circa un anno fa e giunta adesso alla versione 1.0 “production ready”.
Il “TensorFlow Dev Summit” è stato aperto da un keynote di Jeff Dean (wiki, web, lk) - Google Senior Fellow - Rajat Monga (tw) - TensorFlow leader nel Google Brain team - e Megan Kacholia (lk) Engineering Director del TensorFlow/Brain team.

Per approfonsire il #TFDevSummit 2017:

L’incontro del Meetup "Machine-Learning” di Roma è stata un’occasione per rivedere insieme e commentare il video e fare anche una breve presentazione di TensorFlow.
Alla fine dell'evento c’è stato un piccolo rinfresco aperto a tutti gli appassionati di deep learning a Roma.

Simone Scardapane: “TensorFlow and Google, one year of exciting breakthroughs”

Simone (mup, web, fb, lk) ha aperto l’incontro con una breve presentazione sul deep learning e TensorFlow.
“TensorFlow” ha detto Simone “ ha reso accessibili a tutti le reti neurali che sono alla base del deep learning e dell’intelligenza artificiale. Prima di TensorFlow c’era una oggettiva difficoltà nel gestire e allenare reti neurali vaste e complesse”.
Le moderne architetture di deep learning usano infatti reti neurali molto complesse: ad esempio nelle applicazioni di data imaging  i modelli architetturali prevedono decine di milioni di parametri.

170219-reteneurale.jpg

(Rete neurale, immagine tratta da http://joelouismarino.github.io/blog_posts/blog_googlenet_keras.html)

Uno dei grossi vantaggi di TensorFlow è che permette di definire una rete neurale in modo simbolico: lo strumento fornisce inoltre un compilatore efficiente che gestisce in automatico il processo di back-propagation.
TensorFlow può essere utilizzato inoltre con una interfaccia semplificata come Keras arrivando quasi ad una sorta di programmazione dichiarativa.
La prima release di TensorFlow – ha ricordato Simone - è stata rilasciata nel novembre 2015 con licenza aperta Apache 2.0  (cos’è?). Il 15 febbraio 2017 – durante il TFDevSummit - è stata annunciata la versione 1.0 di TensorFlow la prima “production ready”.
La disponibilità di un ambiente deep learning aperto e “user-friendly” ha permesso lo sviluppo di una vasta comunità di esperti, ricercatori e semplici appassionati e il rilascio applicazioni software di grande impatto. Simone ha mostrato alcuni esempi.

1) Neural image captioning: software in grado di riconoscere e descrivere o sottotitolare immagini.

170216-tensorflow-s2.jpg
 

170216-ragazzo-aquilone.jpg

170216-tren-b-n.jpg

2) Google Neural Machine Translation (GNMT)  che ha permesso il rifacimento di “Google translator” grazie al deep learning: invece di tradurre parola per parola ora è possibile analizza il testo nella sua interezza cogliendo significato e il contesto con un livello di accuratezza ormai vicino alla traduzione umana.

nmt-model-fast.jpg

170216-neural-translation.jpg
170216-neural-translation2.jpg

3) Generative Adversarial Networks (GANS) sistemi capaci di generare nuovi dati grazie a un emenorme “training set” e che lavorano con una coppia di reti neurali: la prima produce nuovi dati la seconda controlla la “bontà” del risultato; questi sistemi sono già stati usati per generare immagini artificiali, scenari per video-game, migliorare immagini e riprese video di scarsa qualità.

170216-generative-image

4) Alphago: il deep learning è anche alla base dei recenti spettacolari successi dell’IA nel campo dei giochi da tavolo come la vittoria di Alphago di Google  contro il campione del mondo di GO.

170216-alphago.jpg

5) WaveNet - a generative model for raw audio - capace di generare discorsi che imitano una voce umana con molta più “naturalezza” rispetto ai migliori sistemi di Text-to-Speech oggi esistenti. WaveNet è già stato utilizzato anche per creare musica artificiale.

Simone ha concluso il suo intervento ricordando che di deep learning e ML si parlerà anche in un track specifico alla Data Driven Innovation Roma 2017   che si terrà presso la 3° università di Roma il 24 e 25 febbraio 2017.

Sintesi del keynote di Jeff Dean, Rajat Monga e Megan Kacholia su TensorFlow

Il keynote di apertura del TF DevSummit 2017 condotto da Jeff Dean, Rajat Monga e Megan Kacholia  ha trattato:

  • origini e storia di TensorFlow
  • i progressi da quanto è stata rilasciata la prima versione opensource di TensorFlow
  • la crescente comunità open-source di TensorFlow
  • performance e scalabilityà di TensorFlow
  • applicazioni di TensorFlow
  • exciting announcements!

Jeff Dean

jeff dean.jpg

Jeff ha detto che l’obiettivo di Google con TensorFlow è costruire una “machine learning platform” utilizzabile da chiunque.
TensorFlow è stato rilasciato circa un anno fa ma le attività di Google nel campo del machine learning e deep learning sono iniziati 5 anni fa.
Il primo sistema realizzato – nel 2012 – è stato DistBelief un sistema proprietario di reti neurali adatto ad un ambiente produzione come quello di Google basato su sistemi distribuiti (vedi “Large Scale Distributed Deep Networks” in pdf). DistBelief è stato utilizzato in molti prodotti Google di successo come Google Search, Google Voice Search, advertising, Google Photos, Google Maps, Google Street View, Google Translate, YouTube.
Ma DistBelief aveva molti limiti: “volevamo un sistema molto più flessibile e general purpose” ha detto Jeff “che fosse open source e che venisse adottato e sviluppato da una vasta comunità in tutto il mondo e orientato non solo alla produzione ma anche alla ricerca. Così nel 2016 abbiamo annunciato TensorFlow, una soluzione capace di girare in molteplici ambienti compreso IOS, Android, Raspberry PI, capace di girare su CPU, GPU e TPU, ma anche sul Cloud di Goole ed essere interfacciata da linguaggi come Python, C++, GO, Hasknell, R”.

170216-tensorflow-keynote1.jpg
TensorFlow ha anche sofisticati tool per la visualizzazione dei dati e questo ha facilitato lo sviluppo di una vasta comunità open source intorno a TensorFlow.

170216-tensorflow-keynote2.jpg

Rajat Monga

rajatmonga2.jpg

Rajat Monga ha ufficialmente annunciato il rilascio della versione 1.0 di TensorFlow illustrandone le nuove caratteristiche.

170216-tensorflow-keynote3.jpg

170216-tensorflow-keynote4.jpg

Rajat ha poi illustrato le nuove API di TensorFlow

170218-tf-model.jpg

TensorFlow 1.0 supporta IBM's PowerAI distribution, Movidius Myriad 2 accelerator, Qualcomm SnapDragon Hexagon DSP. Rajat ha annunciato anche la disponibilità di XLA an experimental TensorFlow compiler  specializzato nella compilazione just-in-time e nei calcoli algebrici.

170218-tf-xla.jpg

Megan Kacholia

megankacholia2.jpg

Megan Kacholia ha approfondito il tema delle performancedi TensorFlow 1.0.

170216-tensorflow-keynote7.jpg

In ambiente di produzione si possono utilizzare  molteplici architetture: server farm, CPU-GPU-TPU, server a bassa latenza (come nel mobile) perché TensorFlow 1.0 è ottimizzato per garantire ottime performance in tutti gli ambienti.

170218-tf-performance1.jpg

Megan ha poi illustrato esempi dell’uso di TensorFlow in ricerche d'avanguardia - cutting-edge research – e applicazioni pratiche in ambiente mobile.

170218-tf-research2.jpg

170218-tf-research1.jpg

Conclusione del keynote

In conclusione del Keynote è di nuovo intervenuto Jeff per ringraziare tutti coloro che contribuiscono alla comunità di TensorFlow pur non facendo parte di Google.
“Dalla comunità” ha detto Jeff “arrivano suggerimenti, richieste e anche soluzioni brillanti a cui in Google non avevamo ancora pensato” citando il caso di un agricoltore giapponese che ha sviluppato un’applicazione con TensorFlow su Raspberry PI per riconoscere i cetrioli storti e scartarli nella fase di impacchettamento.
Nel campo della medicina – ha ricordato Jeff – TensorFlow è stato usato per la diagnostica della retinopatia diabetica (qui una sintesi) e all’università di Stanford per la cura del cancro della pelle .

Contatti

•    Meetup “Machine-Learning e Data Science” di Roma: - sito web e pagina Facebook

Approfondimenti

Video

Ulteriori informazioni su TensorFlow

Leggi anche

AG-Vocabolario: 

          Resoconto primo incontro Meetup "Machine-Learning e Data Science" (3 febbraio 2017)   

170203-ml-meetup.jpg

Il 2 febbraio 2017 si è svolto a Roma il primo incontro del Meetup "Machine-Learning e Data-Science"  (fb) presso la Sala Storica di LUISS ENLABS.

Agenda

  • Simone Scardapane e Gabriele Nocco, presentazione del Meetup "Machine-Learning e Data Science"
  • Gianluca Mauro (AI-Academy): Intelligenza artificiale per il tuo business
  • Lorenzo Ridi (Noovle): Serverless Data Architecture at scale on Google Cloud Platform

Simone Scardapane e Gabriele Nocco, presentazione del Meetup "Machine-Learning e Data Science"

161022-simone-scardapane.jpg

(Simone Scardapane)

Simone (mup, web, fb, lk) ha rapidamente illustrato le finalità del Meetup "Machine-Learning e Data-Science" 
“Vogliamo creare una comunità di appassionati e professionisti di ML, AI e Data Science” ha detto Simone, “un luogo dove trovare risposte a domande quali:

  1. Sono appassionato di Ml, dove trovo altri esperti?
  2. Cerchiamo qualcuno esperto di ML, ne conosci?
  3. Mi piacerebbe avvicinarmi a ML, come faccio?”

gabrielenocco.jpeg

(Gabriele Nocco)

Gabriele Nocco (mup , fb, lk) ha annunciato che il secondo evento del Meetup si terrà a Roma il 16 febbraio 2017 al Talent Garden di Cinecittà (mappa) in collaborazione con il Google Dev Group di Roma . Per partecipare occorre registrarsi – gratuitamente – a EventBrite.
“Parleremo di TensorFlow  e proietteremo il keynote ed alcuni momenti salienti del primo TensorFlow Dev Summit per tutti gli appassionati di deep learning e, grazie anche alla gentile sponsorship di Google, avremo il nostro secondo momento di networking condiviso nei bellissimi spazi a nostra disposizione” ha detto Gabriele.
innocenzo-sansone.jpg

(Innocenzo Sansone)

È intervenuto anche Innocenzo Sansone (fb, tw , lk)  – tra gli organizzatori e sponsor – che ha ricordato che il 24 e 25 marzo 2017 a Roma avrà luogo Codemotion  nel quale è previsto – tra gli altri – anche un track specifico sul Machine Learning.

Gianluca Mauro (AI-Academy ): Intelligenza artificiale per il tuo business

170203-gianluca-mauro2.jpg

(Gianluca Mauro)

Gianluca (blog, lk) – ingegnere, imprenditore, esperto di AI, ML e Data Science – è anche uno dei 3 fondatori – insieme a Simone Totaro  (lk)  e Nicolò Valigi (lk) - di AI-Academy  una startup che si prefigge di favorire l’utilizzo dell’Intelligenza Artificiale nei processi di business aziendali (vedi AI Academy manifesto).

ai-academy.jpg

Breve storia dell’Intelligenza Artificiale

Nella prima parte del suo intervento Gianluca ha delineato lo sviluppo storico dell’Intelligenza artificiale.
Gli inizi della IA si devono alla conferenza tenutesi a Dartmouth – USA - nel 1956  ed organizzata da John McCarthy, Marvin Minsky, Nathaniel Rochester e Claude Shannon: per la prima volta si parla di “intelligenza artificiale” e viene indicato l’obiettivo di “costruire una macchina che simuli totalmente l’intelligenza umana” proponendo temi di ricerca che negli anni successivi avranno un grande sviluppo: reti neurali, teoria della computabilità, creatività e elaborazione del linguaggio naturale.

dartmouth.jpg
 
Fonte immagine: http://www.scaruffi.com/mind/ai.html

La ricerca IA viene generosamente finanziata dal governo statunitense fino alla metà degli anni 60: di fronte alla mancanza di risultati concreti i finanziamenti cessano dando origine al primo “AI winter” (1966 – 1980).
Negli anni 80 l’IA riprende vigore grazie a un cambio di paradigma: invece di inseguire l’obiettivo di riprodurre artificialmente l’intera intelligenza umana si ripiega sulla realizzazione di “Sistemi esperti” in grado di simulare le conoscenze in ambiti delimitati.
Anche questo 2° tentativo ha però scarsa fortuna causando il nuovo “AI winter” che si protrae fino agli inizi degli anni 90 quando comincia a imporsi una nuova disciplina: il Machine Learning.

Cos’è il Machine Learning?

AI.jpg

Fonte immagine: http://www.thebluediamondgallery.com/tablet/a/artificial-intelligence.html

Il Machine Learning – ha spiegato Gianluca – è una branca dell’Intelligenza Artificiale che si propone di realizzare algoritmi che a partire dai dati ricevuti in input si adattino in maniera automatica così da produrre risultati “intelligenti” quali previsioni e raccomandazioni.
Gianluca ha fatto l’esempio di un bambino che impara a camminare: non serve conoscere la legge di gravità ed è sufficiente osservare come cammina la mamma e riprovare fino a che non si trova l’equilibrio.

Cos’è il Deep Learning?

Il deep learning è un sottoinsieme del Machine Learning e si rivolge alla progettazione, allo sviluppo, al testing e soprattutto al traning delle reti neurali e di altre tecnologie per l’apprendimento automatico.
Il deep learning è alla base degli spettacolari successi dell’IA nel campo dei giochi da tavolo: la vittoria agli scacchi di Deep Blue di IBM contro il campione del mondo in carica, Garry Kasparov,  e la vittoria di Alphago di Google contro il campione del mondo di GO .

This is the golden age of Artificial Intelligence

Secondo Gianluca Mauro questo è il momento magico per l’IA perché finalmente abbiamo gli strumenti – algoritmi, data, computing power – necessari per realizzare applicazioni di ML a costi sempre più bassi.
Gli algoritmi sono ormai collaudati grazie ai lavori pubblicati negli ultimi anni a cominciare da quelli di Corinna Cortes (“Support-vector networks”) e Davide Rumelhart (“Learning representations by back-propagating errors”).
Il computing power è rappresentato principalmente dalla grande quantità di tecnologie open source a disposizione.
La combinazione di tutti questi fattori è rivoluzionario come dice Chris Dixon, uno dei più noti esponenti del Venture capital USA:
“La maggior parte degli studi di ricerca, degli strumenti e degli strumenti sw legati al ML sono open source. Tutto ciò ha avuto un effetto di democratizzazione che consente a piccole imprese e addirittura a singoli individui di realizzare applicazioni veramente potenti. WhatsApp è stato in grado di costruire un sistema di messaggistica globale che serve 900 milioni di utenti assumendo solo 50 ingegneri rispetto alle migliaia di ingegneri che sono stati necessari per realizzare i precedenti di sistemi di messaggistica. Questo "effetto WhatsApp" sta accadendo adesso nell’Intelligenza Artificiale. Strumenti software come Theano e TensorFlow, in combinazione con i cloud data centers per i training, e con le GPU a basso costo per il deployment consentono adesso a piccole squadre di ingegneri di realizzare sistemi di intelligenza artificiale innovativi e competitivi”.
Secondo Gianluca l’IA presto sarà una necessità per qualsiasi azienda o per citare Pedro Domingos: “A company without Machine Learning can’t keep up with one that uses it”.
Secondo Andrew Ng, chief scientist in Baidu, AI e ML stanno già trasformando le imprese perché le obbligheranno a rivoluzionare i loro processi produttivi così come accadde nell’800 quando fu disponibile per la prima volta elettricità a basso costo (video).
Questo cambiamento culturale è già avvertibile nel Venture Capital e nel Merger & Acquisition: le grandi imprese non cercano solo startup che si occupano di ricerca pura nell’ML ma startup che realizzano servizi e prodotti con ML embedded.
“Siamo all’alba di una nuova era” ha concluso Gianluca “quella del Machine Learning as a feature”.

Lorenzo Ridi (Noovle): Serverless Data Architecture at scale on Google Cloud Platform

lorenzo-ridi.jpeg

(Lorenzo Ridi)

Lorenzo Ridi (mup, fb, lk),  tw) ha presentato un caso d’uso concreto (qui disponibile nella sua versione integrale , anche su SlideShare) per mostrare i vantaggi di usare l’architettura su Google Cloud Platform, attraverso sole componenti serverless, in applicazioni con Machine-Learnin embedded.
Il caso d’uso riguarda una società che con l’avvicinarsi del Black Friday decide di commissionare un’indagine sui social, e in particolare su Twitter, per catturare insights utili a posizionare correttamente i propri prodotti, prima e durante l’evento: questo è tanto più cruciale quanto si considera l’enorme dimensione del catalogo aziendale perché indirizzare in modo sbagliato la propria campagna pubblicitaria e promozionale sarebbe un errore fatale.
Tuttavia, per gestire il forte traffico atteso durante l’evento, gli ingegneri di ACME decidono di abbandonare le tecnologie tradizionali, e di implementare questa architettura su Google Cloud Platform, attraverso sole componenti serverless:

170203-google-architecture-ml1.jpg
 
Ingestion

Per recuperare i dati viene implementata una semplice applicazione Python che, attraverso la libreria TweePy, accede alle Streaming API di Twitter recuperando il flusso di messaggi riguardanti il Black Friday e le tematiche ad esso connesse.
Per fare in modo che anche questa componente mantenga gli standard di affidabilità prefissati, si decide di eseguirla, all’interno di un container Docker, su Google Container Engine, l’implementazione di Kubernetes su Google Cloud Platform. In questo modo, non dovremo preoccuparci di eventuali outage o malfunzionamenti. Tutto è gestito (e all’occorrenza automaticamente riavviato) da Kubernetes.

170203-google-architecture-ml2.jpg
 
Innanzitutto creiamo l’immagine Docker che utilizzeremo per il deploy. A questo scopo è sufficiente redigere opportunamente un Dockerfile che contenga le istruzioni per installare le librerie necessarie, copiare la nostra applicazione ed avviare lo script:

170203-google-architecture-ml3.jpg
 
Et voilà! L’immagine Docker è pronta per essere eseguita ovunque: sul nostro laptop, su un server on-prem o, come nel nostro caso, all’interno di un cluster Kubernetes. Il deploy su Container Engine è semplicissimo, con il tool da riga di comando di Google Cloud Platform: tre sole istruzioni che servono a creare il cluster Kubernetes, acquisire le credenziali di accesso ed eseguire l’applicazione in modo scalabile ed affidabile all’interno di un ReplicationController.
Il secondo elemento della catena, la componente cioè verso la quale la nostra applicazione invierà i tweet, è Google Pub/Sub. una soluzione middleware fully-managed, che realizza un’architettura Publisher/Subscriber in modo affidabile e scalabile.
Nella fase di processing, utilizziamo altri due strumenti della suite Google Cloud Platform:

  • Google Cloud Dataflow è un SDK Java open source – adesso noto sotto il nome di Apache Beam – per la realizzazione di pipeline di processing parallele. Inoltre, Cloud Dataflow è il servizio fully managed operante sull’infrastruttura Google, che esegue in modo ottimizzato pipeline di processing scritte con Apache Beam.
  • Google BigQuery è una soluzione di Analytic Data Warehouse fully managed. Le sue performance strabilianti, che abbiamo avuto modo di sottolineare più volte, lo rendono una soluzione ottimale all’interno di architetture di Data Analytics.

La pipeline che andiamo a progettare è estremamente semplice. Di fatto non farà altro che trasformare la struttura JSON che identifica ogni Tweet, inviata dalle API di Twitter e recapitata da Pub/Sub, in una struttura record BigQuery. Successivamente, attraverso le BigQuery Streaming API, ogni record verrà scritto in una tabella in modo tale che i dati possano essere immediatamente analizzati.
 170203-google-architecture-ml4.jpg
Il codice della pipeline è estremamente semplice; questo è in effetti uno dei punti di forza di Apache Beam rispetto ad altri paradigmi di processing, come MapReduce. Tutto ciò che dobbiamo fare è creare un oggetto di tipo Pipeline e poi applicare ripetutamente il metodo apply() per trasformare i dati in modo opportuno. È interessante osservare come i dati vengano letti e scritti utilizzando due elementi di I/O inclusi nell’SDK: PubSubIO e BigQueryIO. Non è quindi necessario scrivere codice boilerplate per implementare l’integrazione tra i sistemi.

Machine learning

Per visualizzare graficamente i risultati utilizziamo Google Data Studio, uno strumento della suite Google Analytics che consente di costruire visualizzazioni grafiche di vario tipo a partire da diverse sorgenti dati, tra le quali ovviamente figura anche BigQuery.
Possiamo poi condividere le dashboard, oppure renderle pubblicamente accessibili, esattamente come faremmo con un documento Google Drive.

170203-ml5.jpg
 
In questo grafico è riportato il numero di Tweet collezionato da ogni stato dell’Unione. Sicuramente d’impatto, ma non molto utile per il nostro scopo. In effetti, dopo un po’ di analisi esplorativa dei dati, ci accorgiamo che con i soli tweet collezionati non riusciamo a fare analisi molto “avanzate”. Dobbiamo quindi rivedere la nostra procedura di processing per cercare di inferire qualche elemento di conoscenza più “interessante”.
Google Cloud Platform ci viene in aiuto, in questo caso offrendoci una serie di API, basate su algoritmi di Machine Learning, il cui scopo è esattamente aggiungere un pizzico di “intelligenza” al nostro processo di analisi. In particolare utilizzeremo le Natural Language API, che ci saranno utili per recuperare il sentiment di ogni tweet, cioè un indicatore numerico della positività (o negatività) del testo contenuto nel messaggio.

170203-google-architecture-ml6.jpg
 
La API è molto semplice da usare: prende in ingresso un testo (il nostro tweet) e restituisce due parametri:

  • Polarity (FLOAT variabile da -1 a 1) esprime l’umore del testo: valori positivi denotano sentimenti positivi.
  • Magnitude (FLOAT variabile da 0 a +inf) esprime l’intensità del sentimento. Valori più alti denotano sentimenti più forti (siano essi rabbia o gioia).

La nostra personale semplicistica definizione di “sentiment” altro non è che il prodotto di questi due valori. In questo modo siamo in grado di assegnare un valore numerico ad ogni tweet – ed auspicabilmente, di tirarne fuori delle statistiche interessanti!
La pipeline Dataflow viene modificata in modo da includere, oltre al flusso precedente, anche questo nuovo step. Tale modifica è molto semplice, e visto il modello di programmazione di Cloud Dataflow, permette un notevole riuso del codice esistente.

170203-google-architecture-ml7.jpg
 

Con questi nuovi dati possiamo realizzare delle analisi molto più interessanti, che ci informano sulla distribuzione geografica e temporale del “sentimento” riguardante l’evento Black Friday.
La mappa che segue, ad esempio, mostra il sentiment medio registrato in ognuno degli stati degli US, colori più scuri rappresentano sentiment più negativi (quel quadrato rosso là in mezzo è il Wyoming).

170203-google-architecture-ml8.jpg
 
Quest’altra analisi invece riporta l’andamento del sentiment legato ai tre maggiori vendor statunitensi: Amazon, Walmart e Best Buy. A partire da questa semplice analisi, con un po’ di drill-down sui dati, siamo riusciti a carpire alcuni fatti interessanti:

  • il popolo di Twitter non ha apprezzato la decisione di Walmart di anticipare l’apertura delle proprie vendite al giorno precedente il Black Friday, la festa nazionale del Thanksgiving Day. La popolarità di Walmart è stata infatti minata fin dai primi di Novembre da questa decisione  – d’altronde, la tutela dei lavoratori è un tema universale.
  • Le vendite promozionali di Amazon (aperte il 18 Novembre, quindi con anticipo rispetto al Black Friday) sono state inizialmente duramente criticate dagli utenti, con un crollo della popolarità che ha raggiunto il suo minimo il 22. In seguito però il colosso delle vendite online ha recuperato terreno rispetto a Best Buy, che invece sembra aver mantenuto intatta la sua buona reputazione per tutto il periodo.

170203-google-architecture-ml9.jpg

Contatti

Leggi anche

AG-Vocabolario: 

          Simone Scardapane: Machine Learning Lab (1 dicembre 2016)   

161022-ml2.jpg

Il 22 ottobre 2016 si è svolta la “Google DevFest”, organizzata dal Google Developer Group Roma Lazio Abruzzo presso l’Engineering Department dell’Università degli Studi di Roma Tre.
Uno dei seminari della “Google DevFest” è stato dedicato al “Machine Learning” ed è stato condotto da Simone Scardapane - della Università “La Sapienza” di Roma - che abbiamo intervistato.

161022-simone-scardapane.jpg

(Simone Scardapane)
 
Agatino Grillo (AG): Buongiorno Simone e grazie per la collaborazione. Ti vuoi presentare?

Simone Scardapane (SS): Sono un assegnista di ricerca presso l’Università “la Sapienza” di Roma, e le mie attività di ricerca si concentrano in prevalenza su vari aspetti del Machine Learning. Sono membro del GDG da diversi anni, ed in questo tempo abbiamo organizzato diverse presentazioni, come quella dell’ultima DevFest, per cercare di introdurre un pubblico più vasto alle tematiche del Machine Learning, che (credo) saranno sempre più importanti per sviluppatori e non negli anni a venire.
Prima di prendere il dottorato, mi sono laureato in Ingegneria Informatica a Roma Tre, ed in seguito ho conseguito un Master in Intelligenza Artificiale e Robotica presso La Sapienza.

AG: Cos’è il “Machine Learning” (ML)?

SS: Si tratta di un campo molto vasto che, in sostanza, si pone come obiettivo di realizzare algoritmi che si adattino in maniera automatica a partire dai dati che ricevono, al fine di realizzare previsioni o, più in generale, comportamenti intelligenti di vario tipo.
Un esempio classico è una banca che, al momento di erogare un prestito, potrebbe utilizzare un algoritmo di ML per fare previsioni sulla solvibilità dei propri clienti basandosi sullo storico dei prestiti già erogati fino a quel momento.
Un altro campo di utilizzo classico del ML riguarda la “raccomandazione” di prodotti o servizi simili sulla base delle scelte fatte dai consumatori già registrati nelle basi dati, che ha moltissima rilevanza oggi nei servizi di e-commerce.

AG: Come è nato il ML e perché oggi ha acquistato tanta importanza?

SS: Il Machine Learning, che possiamo considerare una branca dell’Intelligenza Artificiale, è una disciplina che sotto varie forme esiste da molti decenni: il primo embrione di quelle che oggi sono le reti neurali nasce negli anni quaranta, mentre il termine “machine learning” viene introdotto circa nel 1959. È un campo complesso che ha guadagnato dal contributo di tantissime discipline, tra cui l’informatica, la matematica, le scienze cognitive e molte altre.
Negli anni ha attraversato varie fasi, direi che oggi ha raggiunto una fase di maturità grazie ad (almeno) 3 fattori concomitanti:

  1. sono disponibili grandi basi di dati che possono essere usate per addestrare gli algoritmi di ML
  2. la potenza computazionale necessaria è disponibile a basso costo (si pensi a piattaforme di cluster computing come Spark)
  3. si sono imposte delle librerie standard che permettono di realizzare gli algoritmi in modo veloce all’interno di linguaggi e programmi estremamente interattivi.

Esiste una classificazione storica degli algoritmi di ML, che possiamo dividere in:

  1. apprendimento supervisionato: si basa sull’utilizzo di informazioni “etichettate” in qualche modo – come nell’esempio della banca, dove ciascun cliente è etichettato in funzione del suo livello di solvibilità – per predire nuovi risultati
  2. apprendimento non supervisionato: in questo caso, l’obiettivo è gestire informazioni secondo strutture non note a priori; ad esempio facendo clustering di clienti per offerte commerciali mirate
  3. apprendimento con rinforzo (reinforcement learning): è l’area più avanzata e probabilmente meno standardizzata, in questo caso i dati servono ad apprendere azioni ottimali in ambienti non strutturati – si pensi ad esempio ad un programma che impara a giocare a scacchi.

AG: Cosa serve per fare ML?

SS: Ovviamente dipende dalla quantità di dati che si vuole analizzare e dagli obiettivi che ci si prefigge. A livello implementativo, se i dati non sono eccessivi si può facilmente sperimentare su un singolo computer, basta la conoscenza di almeno un linguaggio di programmazione e delle librerie di ML e di analisi di dati disponibili. A livello teorico conta molto quanto si vuole approfondire l’argomento. Alcuni concetti di fondo si imparano facilmente, ma per andare oltre le descrizioni informali dei metodi e degli algoritmi è richiesta (almeno) una conoscenza di base di algebra lineare e di ottimizzazione.

161022-ml3.jpg

AG: in cosa è consistito il tuo ML lab?

SS: Nel lab che ho condotto durante la Google Fest abbiamo usato scikit-learn (sklearn) - una libreria open source di apprendimento automatico in Python. L’idea era quella di mostrare che, con il livello di maturità raggiunto dalle librerie software di oggi, si può arrivare a dei primi risultati in maniera rapida e valutando tutto il processo ad alto livello. Ovviamente questo non toglie la necessità di capire gli algoritmi, ma credo che questo genere di esperienza “hands-on” aiuti ad invogliare le persone ad avvicinarsi a questo campo.

Nel dettaglio, sklearn contiene algoritmi ML di classificazione, regressione, clustering, e diverse funzioni di supporto per la valutazione e l’ottimizzazione dei modelli. Soprattutto, ha un’interfaccia molto semplice e consistente per il suo utilizzo. Per chi volesse replicare quanto fatto, consiglio di scaricare e installare l’ambiente Pyhton integrato Anaconda, un full-stack che contiene tutto quello che serve, incluso un IDE specializzato per il calcolo scientifico.

Nel lab abbiamo visto e commentato insieme tre esempi di algoritmi ML:

  • Music classification: classificare secondo il genere musicale un insieme di canzoni tratte da un dataset predefinito.
  • Image segmentation effettuata in maniera non convenzionale grazie a un algoritmo di clustering.
  • Joke recommendation: come consigliare barzellette usando il dataset di prova 1_1 della Jester collection.

Tutto il codice zippato e le slides relative sono disponibili sulla mia pagina web.

AG: Che sviluppi futuri ti aspetti per il ML?

SS: Negli ultimi due/tre anni abbiamo visto una esplosione di applicazioni grazie alla diffusione del “deep learning”, una nuova famiglia di algoritmi che permettono di estrarre rappresentazioni gerarchiche di dati complessi (es., immagini). Sicuramente questo trend continuerà, ad esempio con nuovi filoni di applicazioni in ambito biomedico e di analisi di dati scientifici (per esempio nel campo della fisica delle particelle).
Più in generale, tutti i risultati ottenuti finora sono solo una piccola parte di quello che servirebbe per avere degli algoritmi che “imparano” nel senso più generale del termine. L’accelerazione del ML e del deep learning sicuramente porterà ad una simile accelerazione in questi campi più di frontiera, soprattutto nel reinforcement learning, nell’analisi di dati complessi, nel trasferimento di conoscenza fra problemi diversi, e così via.

AG: Cosa consigli a chi voglia intraprendere questo tipo di studi a livello universitario?

SS: Come già detto, il ML è un campo molto variegato, a cui ci si può affacciare da diverse discipline, incluse informatica, ingegneria, matematica e così via. Il mio consiglio per chi volesse specializzarsi in questa disciplina è di controllare bene quale sia l’opzione migliore all’interno della propria università – in alcuni casi potrebbe essere preferibile un percorso all’interno di Informatica o Ingegneria Informatica, in altri i corsi potrebbero essere erogati all’interno dell’area di matematica applicata o statistica. È difficile dare un consiglio che valga in generale.

AG: Cosa consigli invece a chi voglia approfondire l’argomento?

SS: Oggi il modo più comune di approcciarsi al ML è quello di frequentare uno dei tanti corsi online disponibili su piattaforme MOOC, come quelli di Coursera o Udacity. Si tratta di ottimi punti di partenza che permettono un buon ripasso dei fondamenti ed una certa visione di insieme. A quel punto è possibile seguire materiale più specializzato a seconda dei propri interessi. Il mio consiglio in questo caso è di provare da subito ad applicare le tecniche che si imparano, ad esempio all’interno di piattaforme come Kaggle che mettono a disposizione diversi dataset su cui confrontarsi. Come in tutti i settori, l’esperienza pratica non è mai troppo poca, ed è l’unica che permette di prendere realtà familiarità con una vasta schiera di tecniche ed algoritmi.

Link

Google Developer Group Lazio e Abruzzo

Simone Scardapane

Strumenti e tool

 

AG-Vocabolario: