The many forms of Artificial Intelligence

The many forms of Artificial Intelligence

AI is an umbrella term for a range of computer algorithms and approaches that allow machines to sense, reason, act and adapt like humans. Machine Learning AI encompasses a whole set of different computing methods, a major subset of which is called “machine learning.” As Intel’s Dubey explains it, machine learning “is a program where performance improves over time,” and that also gets better with more data input. A more formal definition of machine learning used at Intel is: “the construction and study of algorithms that can learn from data to make predictions or decisions.” Wired magazine declared “the end of code” in describing how machine learning is changing programming: “In traditional programming, an engineer writes explicit, step-by-step instructions for the computer to follow. They train them.” Using machine learning, a major eye hospital in China was able to improve detection of potential causes of blindness, typically 70 to 80 per cent for clinicians, to 93 per cent. Most simply, deep learning is a specific method of machine learning, and it’s based primarily on the use of neural networks. “In traditional supervised machine learning, systems require an expert to use his or her domain knowledge to specify the information (called features) in the input data that will best lead to a well-trained system,” wrote a team of Intel AI engineers and data scientists in a recent blog. Deep learning is different. “Rather than specifying the features in our data that we think will lead to the best classification accuracy,” they continued, “we let the machine find this information on its own. Often, it is able to look at the problem in a way that even an expert wouldn’t have been able to imagine.” The neural network – technically an “artificial neural network” since it’s based on how we think the brain works – provides the math that makes it work. The important part is this: The neural network allows the program to break a problem down into smaller and smaller – and therefore simpler and simpler – chunks. “Deep” in deep learning delineates the use of a many-layered neural network. With more layers, the program gets more refined in what it can categorize and more accurate in doing so – it just requires more and more data and more and more computing power. Training is the part of machine learning in which you’re building your algorithm, shaping it with data to do what you want it to do. “Training is the process by which our system finds patterns in data,” wrote the Intel AI team. “During training, we pass data through the neural network, error-correct after each sample and iterate until the best network parametrization is achieved. After the network has been trained, the resulting architecture can be used for inference.” And then there’s inference, which fits its dictionary definition to the letter: “The act or process of deriving logical conclusions from premises known or assumed to be true.” In the software analogy, training is writing the program, while inference is using it. “Inference is the process of using the trained model to make predictions about data we have not previously seen,” wrote those savvy Intel folks. It uses algorithms based on neural networks – a way to connect inputs and outputs based on a model of how we think the brain works – that find the best way to solve problems by themselves, as opposed to by the programmer or scientist writing them. In general, AI is an umbrella term for a range of computer algorithms and approaches that allow machines to sense, reason, act and adapt like humans do – or in ways beyond our abilities.

Read More

12 Frequently Asked Questions on Deep Learning

12 Frequently Asked Questions on Deep Learning

From Facebook’s research to DeepMind’s legendary algorithms, deep learning has climbed its way to the top of the data science world. Having said that, a PhD in a specific field (like linguistics for NLP) will definitely accelerate your path if you choose to combine that with deep learning. The python ecosystem comprises of developers and coders who are providing open source libraries and support for the community of python users. This makes the task of writing complex codes for various algorithms much easier and the techniques easier to implement and experiment with. Also, Python being a more generalized programming language, can be used for both the development and implementation. That is, a deep learning product that can predict the price of flight tickets, can not only be developed in python but can also be attached with your website in the same form. This makes experimentation easier by providing abstraction to the unnecessary information that is hidden under the algorithms. And giving access to the parameters that can be tweaked to enhance the performance of such models. When you press the buttons on a television remote, do you need to care about the background processes that are happening inside the remote? Do you need to know about what signal is being sent out for that key, or how is it being amplified? Because maybe an understanding of these processes is required for a physicist but for a lame man sitting in his bedroom, it is just an information overload. There are also other contenders apart from Python in the deep learning space such as R, Julia, C++, and Java. If you are not well versed with programming, there are also a few GUI based softwares, that require no coding, to build deep learning models, such as Lobe or Google’s AutoML, among others. In forward pass, input is passed through the neural network and after processing the input, an output is generated. Whereas in backward pass, we update the weights of neural network on the basis of error we get in forward pass. Here, we can see that each element in one row of first array is multiplied with one column of second array. So in a neural network, we can consider first array as input to the neural network, and the second array can be considered as weights of the network. Now just to give you a sense of what kind of scale deep learning – VGG16 (a convolutional neural network of 16 hidden layers which is frequently used in deep learning applications) has ~140 million parameters; aka weights and biases. Now think of all the matrix multiplications you would have to do to pass just one input to this network! We saw that the computationally intensive part of neural network is made up of multiple matrix multiplications. We can simply do this by performing all the operations at the same time instead of doing it one after the other. This, in a nutshell, is why we use GPU (graphics processing units) instead of a CPU (central processing unit) for training a neural network. Deep Learning have been in the spotlight for quite some time now. Its “deeper” versions are making tremendous breakthroughs in many fields such as image recognition, speech and natural language processing etc. Now that we know it is so impactful; the main question that arises is when to and when not to apply neural networks? And to be a part of this “gold rush”, you have to keep a few things in mind: Firstly, deep learning models require clear and informative data (and mostly big data) to train. Try to imagine deep learning model as a child. Then it tries to walk on its own, and with its every step, the child learns how to perform a particular task. It is prudent to use Deep Learning for complex problems such as image processing. Deep Learning algorithms belong to a class of algorithms called representation learning algorithms. When you have an appropriate type of deep learning to solve the problem. Each problem has its own twists. So the data decides the way you solve the problem. Whereas, if it is image related problem, you would probably be better of taking convolutional neural networks for a change. Last but not the least, hardware requirements are essential for running a deep neural network model. Neural nets were “discovered” long ago, but they are shining in the recent years for the main reason that computational resources are better and more powerful. Do we need a lot of data to train deep learning models? It is true that we need a large amount of data to train a typical deep learning model. One of the barrier for using deep learning models for industry applications is where the data is not in huge amount. A few examples of data needed to train some of the popular deep learning models are: However, a deep learning model trained on a specific task can be reused for different problem in the same domain even if the amount of data is not that huge. For instance, we have a set of 1000 images of cats and dogs labeled as 1 and 0 (1 for cat and 0 for dog) and we have another set of 500 test images that we need to classify. So, instead of training a deep learning model on the data of 1000 images, we can use a pre-trained VGGNet model and retrain it on our data and use it to classify the unlabeled set of images. You may have a look at this article to get a better intuition of using a pre-trained model. For intermediate users, this Age Detection challenge is a nice project to work on. The dataset consists of facial images of Indian movie actors. The task is to predict the age of a person from his or her facial attributes. For simplicity, the problem has been converted to a multiclass problem with classes as Young, Middle and Old. Do we need a lot of data to train deep learning models? What are some of the free learning resources for Deep Learning? Being a comparatively newer technology, there is not enough content and tutorials available for the beginners. The learning resources can be classified on the different applications of deep learning. Besides this, you can also go through the following blogs for a more extensive list of resources: What are some Deep Learning interview questions? Do note that this is not an exhaustive list that will make you completely ready for an interview. You can go through the following skill test to test yourself on important questions on deep learning. However if you are a newcomer to this field, the word “deep” might throw you into doubt. Deep learning has come a long way in recent years, but still has a lot of untapped potential. One of the use-cases that we can definitely see in the suture is of automobile industry, where Deep Learning can revolutionize it by making self-driving cars a reality. While we don’t have a crystal ball to predict the future, we can see deep learning models requiring less and less involvement from human data scientists and researchers. In the immediate future, we can definitely see a trend where the knowledge of deep learning will be a skill required by every Data Science practitioner. This person is responsible to deploy and maintain Deep Learning models used by various departments of that company. Needless to say, there will be a huge demand of such people in the industry. Currently, one of the limitations of DL is that it does what a human asks of it. It requires tons of data to learn it’s target objective, and replicates that. We can see this improving over time such that the bias is eliminated in the training process. We might even stop differentiating deep learning from the other types of learning, with time. It is primed to become a popular and commonly used field and will not require special branding efforts to market or sell. There are a lot of cases still where researchers, after training a DL model, are unable to explain the ‘why’ behind it. “It’s producing great results but why did you tune a hyperparameter a certain way?” Hopefully with the rapid advancement in DL, we will see this black box concept becoming history, and we can explain the intuition behind the decision it takes. These are broadly the answers to the most frequently asked questions on our portal or elsewhere by the people who want to jump onto this Deep Learning bandwagon. Do you have any other questions on deep learning that you need clarification on? Use the comments section below to ask; or hop onto our Discussion portal, and we will help you out! Deep Learning is nothing but a paradigm of machine learning which has shown incredible promise in the recent years. This is because of the fact that Deep Learning shows great analogy with the functioning of the human brain. The superiority of the human brain is an evident fact, and it is considered to be the most versatile and efficient self-learning model that has ever been created. Let us understand the functioning of a deep learning model with an example: Deep learning is one of the hottest topics of this industry today, but it is unfortunately foreign and cryptic to most people. Despite the fact, that there is sand, greenery, clouds and a lot of other things, our brain tags this image as one of a car. This is because our brain has learnt to identify the primary subject of an image. This ability of deriving useful information from a lot of extraneous data is what makes deep learning special. Now although Deep Learning has been around for many years, the major breakthroughs from these techniques came just in the recent years. This is because of two main reasons – the first and foremost, as we saw before, is the increase of data generated through various sources. GPUs, which are becoming a requirement to run deep learning models, are multiple times faster and they help us build bigger and deeper deep learning models in comparatively less time than we required previously. This is the reason that Deep Learning has become a major buzz word in the data science industry. A lot of people carry an impression that deep learning involves a lot of mathematics and statistical knowledge. From Netflix’s famous movie recommendation system to Google’s self-driving cars, deep learning is already transforming a lot of businesses and is expected to bring about a revolution in almost all industries. Deep learning models are being used from diagnosing cancer to winning presidential elections, from creating art and writing literature to making real life money. Thus it would be wrong to say that it is just a hyped topic anymore. This is being done through some deep learning models being applied to NLP tasks and is a major success story. Voice commands have added a whole new domain to the possibilities of a machine. If you had similar questions about deep learning, but were not sure how, when and where to ask them – you are at the right place. Deep Learning is being used in Healthcare domain to locate malignant cells and other foreign bodies in order to detect complex diseases. However, some people develop a thinking that deep learning is overhyped because of the fact that labeled data required for training deep learning models is not readily available. Even if the data is available, the computational power required to train such models does not come cheap. Hence, due to these barriers, people are not able to experience the power of deep learning and term it as just hype. This is one of the most important questions that most of us need to understand. The most important difference between deep learning and traditional machine learning is its performance as the scale of data increases. When the data is small, deep learning algorithms don’t perform that well. This is because deep learning algorithms need a large amount of data to understand it perfectly. Feature engineering is a process of putting domain knowledge into the creation of feature extractors to reduce the complexity of the data and make patterns more visible to learning algorithms to work. This process is difficult and expensive in terms of time and expertise. In Machine learning, most of the applied features need to be identified by an expert and then hand-coded as per the domain and data type. The performance of most of the Machine Learning algorithm depends on how accurately the features are identified and extracted. This is a very distinctive part of Deep Learning and a major step ahead of traditional Machine Learning. By end of this article, we will dispel a few myths about deep learning and answer some widely asked questions about this field. Therefore, deep learning reduces the task of developing new feature extractor for every problem. Like, Convolutional NN will try to learn low-level features such as edges and lines in early layers then parts of faces of people and then high-level representation of a face. Last but not the least, we have interpretability as a factor for comparison of machine learning and deep learning. This factor is the main reason deep learning is still thought 10 times before its use in industry. Indeed mathematically you can find out which nodes of a deep neural network were activated, but we don’t know what there neurons were supposed to model and what these layers of neurons were doing collectively. On the other hand, machine learning algorithms like decision trees give us crisp rules as to why it chose what it chose, so it is particularly easy to interpret the reasoning behind it. Therefore, algorithms like decision trees and linear/logistic regression are primarily used in industry for interpretability. If you would like to learn about a more in-depth comparison between machine learning and deep learning, I recommend you go through the following blog: What are the prerequisites for starting out in Deep Learning? Understanding the concepts of statistics are essential because most of the deep learning concepts are derived from assimilating the concepts of statistics. Here is the exciting part – It isn’t as difficult as most people make it out to be. Read on to find more! Coursera’s, Introduction to Data Science in Python is a decent course to start off with Python as a tool. One can not start learning deep learning without understanding the concepts of machine learning. Do I need to do a PhD to make a career in Deep Learning? No, a PhD is not a mandatory requirement to make a career in deep learning.

Read More

Stock Prediction Algorithm: FB, AAPL, MU & AMZN + Building A Dynamic Portfolio With Artificial Neural Networks

Stock Prediction Algorithm: FB, AAPL, MU & AMZN + Building A Dynamic Portfolio With Artificial Neural Networks

Apple only announced its $100 billion share buyback program after it did its Q2 earnings report last May 1. I do not have any real evidence but Berkshire Hathaway bought 75 million AAPL shares in the first quarter of this year because it already got a hint from Apple’s management that it will allocate $100 million for share buybacks. The aim of the conference was to develop two-way learning connections between China and Israel and to familiarize the Chinese firms with the rapidly changing technology coming out of Israel, and to network between firms from the two countries. For more of a background in explaining this conference: SITI is a leading dual-side cooperation system platform in investing in potential start-up companies in both China and Israel. AI is rapidly developing and as we have mentioned China is at the forefront of this development. Rapid developments by China's AI leaders Baidu, Alibaba, and Tencent, we're quickly looking at another technology revolution, and researchers at the University of Toronto have noted that China is becoming an AI superpower faster than the world has anticipated. The researchers noted, "Innovation is not a zero-sum game […] AI technologies developed abroad will inevitably be used and further developed [elsewhere]." This is an exciting opportunity to interact closely with the center of rapid growth of AI and expand our opportunities. You should also know:Recently, the dollar has had a prosperous run and hit a 2018 high, pushing the bullion lower. The Federal Reserve meeting which was discussed last week has not had much of an impact as the results of the meeting were much as expected, and Federal Reserve Chair Jerome Powell’s statements remained fairly unchanged since March. According to the Fed, inflation has moved close to its 2% target, a number which stirs up much debate as to its appropriateness. Currently, this is the longest and tightest (in terms of percentages) gold price consolidation since the bottom in December 2015. Jens Pederson, a Danske Bank senior analyst, stated, “A stronger dollar has created headwinds for gold but we don’t see the dollar going much higher on a medium-term basis and in terms of geopolitics there are some factors to keep an eye on”. The dollar’s retreat is in light of the latest news that President Trump might withdraw from the Iran nuclear deal. This news will likely push gold higher as worries of the geopolitical consequences ripple through the market. Immediately following the announcement, the gold market has shown signs of volatility and the prices have been fluctuating rapidly within a short range. In the longer term, gold is expected to deliver its strongest annual price performance in five years in 2018, according to GFMS analysts. They believe that the political uncertainty of late will drive investment in bars and bullion-backed investment funds. In the short-run analysts believe that the level of $1300 and $1307 will hold and gold will not drop back down to $1200 levels. I am sure Mr. Buffett will agree that there are long-term benefits to fulfilling the needs of long-time Apple Mac Mini and iPad Mini customers. You should also know:Aluminum is a key material in many of Apple’s most popular products, and for more than 130 years, it’s been produced the same way. Aluminum giants Alcoa Corporation and Rio Tinto Aluminum announced a joint project to commercialize original technology that eliminates direct greenhouse gas emissions from the traditional smelting process. This is a key step in aluminum production that if fully developed and implemented, will strengthen the closely integrated Canada-United States aluminum and manufacturing industries. As part of Apple’s commitment to reducing the environmental impact of its products through innovation, the company helped accelerate the development of this technology. Apple has partnered with both aluminum companies, and the Governments of Canada and Quebec, to collectively invest a combined $144 million to future R&D. “Apple is committed to advancing technologies that are good for the planet and help protect it for generations to come,” said Tim Cook, Apple’s CEO. This follows Apple’s announcement last month that all of its facilities are now powered with 100 percent clean energy and 23 of its suppliers have committed to do the same. There’s long-term reward in buying back the loyalty and spending power of frustrated Mac Mini and iPad Mini users who bought other brands because Apple is very tardy in upgrading its small form Mac computer and iOS tablet. Apple achieved a quarterly revenue of $61.1 billion, 16% increase from Q2 of 2017, quarterly earnings per diluted share of $2.73, up 30%, and generated over $15 billion in operating cash flow. In Q2, iPhone X was sold more than any other iPhone each week and the company’s revenue in all geographic segments grew, with over 20% growth in Greater China and Japan. Reflecting the approved increase, the board has also declared a cash dividend of $0.73 per share of Apple’s common stock. From the inception of its capital return program in August 2012 through March 2018, Apple has returned $275 billion to shareholders, including $200 billion in share repurchases. The Company will complete the execution of the previous $210 billion share repurchase authorization during Q3. The late Steve Jobs himself was the proud presenter of the first Mac Mini small form factor Mac computer in 2005. Apple offered the first version of the Mac Mini for just $499. This is in spite of the stock’s substantial gain from $9.55 last April 3 to $12.8 this week. This was days after I made by buy recommendation for the stock last March 29.That March article was to dispel the downgrade-to-sell assessment by Susquehanna analyst Christopher Rolland for AMD. He finally realized that AMD’s tailwind from GPU-using Ethereum miners will remain strong in spite of the upcoming release of ASIC mining hardware from Bitmain. The commitment of Ethereum and Monero blockchain developers to fork their cryptocurrency algorithm as an anti-ASIC move truly fortified the long case for AMD. It means I Know First has an impeccable accuracy in predicting the short and long-term market movement of AMD. After adjusting for stock-based compensation and other factors, the company claimed earnings of 11 cents a share, up from a break-even performance a year ago. The solid growth came from the division responsible for central processing units designed for personal computers and GPUs, with AMD reporting that average selling prices for both GPUs and CPUs increased year-over-year and quarter-over-quarter. That segment collected revenue of $1.12 billion, destroying the average analyst estimate of $926 million with a year-over-year growth rate of 95%. AMD continued to return innovation and excitement to the PC market with the introduction of new consumer-focused desktop processors. AMD introduced the first AMD Ryzen desktop APUs, combining the high-performance Radeon “Vega” graphics architecture with revolutionary “Zen” CPU cores on a single chip. AMD also delivered the next products in the Company’s strong multi-generational roadmap with the launch of its 2nd Generation Ryzen desktop CPUs just over one year after bringing the first Ryzen processors to market. The new 2nd Generation Ryzen processors can deliver up to 15% higher gaming performance compared to 1st Generation Ryzen processors, with the Ryzen 7 2700X processor delivering the highest multiprocessing performance available on a mainstream desktop PC. Adoption of AMD products for the datacenter continued with new AMD EPYC processor-powered platforms and deployments. AMD brought the powerful “Zen” architecture to a variety of new embedded markets with the launch of EPYC Embedded and Ryzen Embedded processors. The reason for AAPL’s quick rise to above $188 is due to Warren Buffett’s May 3 evening revelation that his Berkshire Hathaway team bought 75 million more Apple shares in Q1 2018. AMD released new Radeon Software Adrenalin Edition updates to optimize the performance of some of the most popular PC games and improve the eSports experience for Radeon graphics card owners. AMD and Microsoft announced support for Radeon FreeSync technology in Microsoft’s Xbox One S and Xbox One X consoles, bringing the exceptional tear-free gaming experience to a broader base of gamers. Over the past 12 months, Micron Technologies stock has risen from its original price of $28.90 to $51.80, jumping a whopping 79%. The following explanations give rise to the rapid increase in MU stock despite the industries characteristic volatile movements. Micron Technology underperformed the industry-wide Server DRAM market in 2017 despite forecasts of strong growth in the sector. Additionally, internet protocol traffic in public clouds data centers is expected to nearly triple from 96,054 Petabytes per month in 2016 to 278,108 PB per month in 2021. The charts aren’t the only thing suggesting shares will continue to rise, because analysts have been aggressively upping their price targets on the stock. Since March 8, analysts have raised the average price target on the stock by nearly 21.5% to about $71.50, according to data from YCharts. Of the 30 analysts covering the stock, about 80% of them rate the shares either a buy or outperform. Additionally, Micron found a stable level of technical support at $45.25 at the start of May on two separate occasions, another bullish technical pattern, known as a double bottom. Additionally, the RSI is now beginning to trend higher, and with a reading around $60, it still has a distance to go before reaching overbought conditions around $70. AAPL’s closing price rose from $176.21 in May 3 to $188 by May 11. AAPL even hit a new 52-week high of $190.37 on May 10. I was correct to claim that the $100 billion stock repurchase plan of Apple was aphrodisiac to long-term value investors like Warren Buffett. Read more.ORLY: It's Time To Invest Like Warren BuffettO’Reilly Automotive, Inc. (NASDAQ: ORLY), together with its subsidiaries, engages in the retail of automotive aftermarket parts, tools, supplies, equipment, and accessories in the United States. Its stores provide do-it-yourself and professional service provider customers a selection of brand name, house brands, and private label products for domestic and imported automobiles, vans, and trucks. In the light of the Berkshire Hathaway Annual Shareholder meeting on Saturday, May 5, 2018 the financial analysts’ community discussed the potential investments targets for the Warren Buffet’s fund. As such, O’Reilly appears to fit into the Buffet’s rules framework and was rated as second stock in the targets list. As of now, the top institutional holders of ORLY are Blackrock and Vanguard Group funds which is itself a positive company of investors and should Warren Buffet join this club, the growth momentum for ORLY’s stock will only improve. Over the last year ORLY had a price gain of +4.45%, YTD gain of +8.17% and P/E ratio of 21.28 based on TTM financials. In comparison with the retail industry in general we can see that the market fell significantly over TTM which is 62.93%. at the same time the industry average for P/E is 43.04 based on TTM. Warren Buffett or his Berkshire Hathaway obviously suspected prior to the Q2 FY 2018 earnings report that Apple will allocate a massive portion of its cash from abroad for stock buybacks. That said, it brings us to ORLY outperforming the falling market with significant difference, while presenting less overvaluation by the market and hence expectancy of the future growth. On May 8, 2018 the company announced two important points: (1) appointment of new CEO and COO, and (2) increase in the maximum target leverage ratio from 2.25 to 2.5. The first matter was expected by the market as the company announced the leadership succession on February 7, 2018. The expectations are that this measure should have positive impact on the overall ORLY performance and bring the company to a better sustainability rank which is also extremely important for the company’s stakeholders. Based on the data from April 2018, the company is bouncing around average values for sustainability indicators among its peer companies in the retail industry. However, we can see that the company has relatively low ranking from governance aspect which is crucial for further company’s growth, maintaining positive reputation and business relationships with O’Reilly Automotive’s stakeholders. Dear Readers,We like to keep you updated on the ongoings within the company and what we are doing to expand our reach and capabilities. Recently, we were invited to the China-Israel Scientific and Technological Innovation Summit in Shenzhen, China to present the company and speak about artificial intelligence and machine learning algorithms. The conference was a multi-day event and we participated in the forums on May 16th.

Read More

Mueller issues grand jury subpoenas to Trump advisor’s social media consultant

Mueller issues grand jury subpoenas to Trump advisor’s social media consultant

U.S. Justice Department Special Counsel Robert Mueller has issued two subpoenas to a social media expert who worked for longtime Donald Trump advisor Roger Stone during the 2016 presidential election campaign. The subpoenas were delivered late last week to lawyers representing Jason Sullivan, a social media and Twitter specialist Stone hired to work for an independent political action committee he set up to support Trump, Knut Johnson, a lawyer for Sullivan, told Reuters on Tuesday. For example, at 6:43 a.m. local time on Election Day in 2016, Trump tweeted, "TODAY WE MAKE AMERICA GREAT AGAIN." Trump's message soon was retweeted more than 343,000 times, and in an interview last year, Sullivan told Reuters that the swarm helped overcome a surge in pro-Clinton social media postings and boost voter turnout for Trump. Stone on Tuesday repeated his public denials that he had an inside track to WikiLeaks or others who hacked or published Democratic Party and Clinton-related emails and said no one from Mueller's team has tried to contact him. The subpoenas suggest that Mueller, who is probing Russian meddling in the 2016 U.S. presidential election, is focusing in part on Stone and whether he might have had advance knowledge of material allegedly hacked by Russian intelligence and sent to WikiLeaks founder Julian Assange, who published it. House of Representatives Intelligence Committee last September and denied allegations of collusion between the president's associates and Russia during the election. "I am aware of no evidence whatsoever of collusion by the Russian state or anyone in the Trump campaign," Stone told reporters at the time. According to sources familiar with the ongoing investigation, Mueller also has been probing whether anyone associated with the Trump campaign may have helped Assange or the Russians time or target the release of hacked emails and other social media promoting Trump or critical of Democratic candidate Hillary Clinton. Sullivan told Reuters that he heads Cyphoon.com, a social media firm, and "worked on the Trump campaign serving as Chief Strategist directly to Roger J. Stone Jr."

Read More

Apple and Its Rivals Bet Their Futures on These Men’s Dreams – Bloomberg

Apple and Its Rivals Bet Their Futures on These Men’s Dreams – Bloomberg

Computers have learned to recognize faces and objects, understand the spoken word, and translate scores of languages. This is the peculiar story—pieced together from my interviews with them—of why it took so long for neural nets to work, how these scientists stuck together, and why Canada, of all places, ended up as the staging ground for the rise of the machines.(Now, not everyone agrees with Canada’s pride of place. That will take decades.BENGIO: I don’t think that humans will necessarily be out of jobs, even if machines become very smart and maybe even smarter than us. We’ll always want real people for jobs that really are about human interactions. I believe that if we’re able to build machines that are as smart as us, they will also be smart enough to understand our values and our moral system, and so act in a way that’s good for us.My real concern is around the potential misuse of AI, for example as applied to military weapons. We need to become collectively wiser.SUTTON: I think it’s a big mistake that we’ve called the field “artificial intelligence.” It makes it seem like it’s very different from people and like it’s not real intelligence. It makes people think of it as more alien than it should be, but it’s a very human thing we’re trying to do: re-create human intelligence.Science has always revealed truths that not all people like—you get the truth but not always the one you wanted. We shouldn’t want to freeze the way we are now and say that’s the way it should always be.Hinton (second from left) and Bengio (right) outside London at a workshop organized by the Gatsby Institute in 2011.LECUN: Until we know exactly what it’s going to look like, worrying about this really is premature. I don’t believe in the concept of singularity, where one day we’ll figure out how to build superintelligent machines and the next day that machine will build even smarter ones and then it will take off. I think people forget that every physical or social phenomenon will face friction, and so an exponentially growing process cannot grow indefinitely.This Hollywood scenario where some genius somewhere in Alaska comes up with the secret to AI and builds one robot and it takes over the world, that’s just preposterous.TRUDEAU: It’s not something I overly fret about. I’m reassured that Canada is part of it in terms of trying to set us on the right path. And I wouldn’t want to slow down our research, our trying to figure out the nuts and bolts of the universe.The question is: What kind of world do we want? Do we want a world where the successful have to hide behind gated communities and everyone else is jealous and shows up with pitchforks? Or do you want a world where everyone has the potential to contribute to innovation?HINTON: I think the social impact of all this stuff is very much up to the political system we’re in. Intrinsically, making the production of goods more efficient ought to increase the general good. The only way that’s going to be bad is if you have a society that takes all of the benefit of that rise in productivity and gives it to the top 1 percent. One of the reasons I live in Canada is its tax system; if you make a lot of money, the country taxes you a lot. It could be driving a simulated car down a road or trying to recognize a cat in a photo.Within that, there’s a subset of machine learning called deep learning. The general idea is you build a neural network, and it has weights and biases that can be tweaked to home in on the desired outcome. That’s what Geoff Hinton and others have really worked on over the past decades, and it’s now the underpinning of what’s most exciting about AI. It does a better job of mimicking the way a human brain thinks.Featured in Bloomberg Businessweek, May 21, 2018. Subscribe now.CADE METZ, reporter for the New York Times and author of a forthcoming history of AI: The idea of a neural network dates back to the 1940s—the notion of a computing system that would mimic the web of neurons in the brain. Navy and other parts of the government, and he developed this thing called a Perceptron based off the neural network concept. When he revealed it, places like the New York Times and the New Yorker covered it in pretty grand terms.Rosenblatt claimed it would not only learn to do small tasks like recognize images but also could theoretically teach machines to walk and to talk and to show emotion. But it was a single layer of neurons, and that meant it was extremely limited in what it could do. Needless to say, none of the things he promised actually happened.Marvin Minsky, a colleague of Rosenblatt’s who happened to be one of his old high school classmates from the Bronx, wrote a book in the late 1960s that detailed the limitations of the Perceptron and neural networks, and it kind of put the whole area of research into a deep freeze for a good 10 years at least.GEOFF HINTON: Rosenblatt’s Perceptron could do some interesting things, but he got ahead of himself by about 50 years. The book by Minsky and Seymour Papert on the technology (Perceptrons: An Introduction to Computational Geometry) basically led to the demise of the field.During the 1970s a small group of people kept working on neural nets, but overall we were in the midst of an AI winter. METZ: Geoff Hinton, at Carnegie Mellon University and then later at the University of Toronto, stuck with the neural network idea. Eventually he and his collaborators and others developed a multilayered neural network—a deep neural network—and this started to work in a lot of ways.A French computer scientist, Yann LeCun, spent a year doing postdoctoral research at Hinton’s lab in Toronto. I grew up in the 1960s, so there was space exploration, the emergence of the first computers, and AI. So when I started studying engineering, I was really interested in artificial intelligence, a field that was very nascent.LeCun (right) at Esiee ​​​​Paris graduate school in 1979.I heard about the Perceptron and was intrigued, because I thought learning was an integral part of intelligence. As an engineer, if you want to understand intelligence, the obvious approach is to try to build a smart machine—it forces you to focus on the components needed to foster intelligence. You don’t want to just mimic biological intelligence or the brain, because there are a lot of aspects of its function that are just due to biochemistry and biology—they’re not relevant to intelligence, really. Like how feathers aren’t crucial for flight: What’s important are the underlying aerodynamic principles.METZ: There were people who thought LeCun was a complete nut and that this was sort of a Sisyphean task. You would go to these big AI conferences as a neural network researcher, and you weren’t accepted by the core of academia. I had a scholarship from the government, so I could basically choose my topic, and it didn’t cost anything to the professor. We made a deal that I could do machine learning, but I would apply it to the thing that he cared about, which was speech recognition.LECUN: Around 1986, there was a period of elation around neural nets, partly due to the interest in those models from physicists who came up with new mathematical techniques. That made the field acceptable again, and this led to a lot of excitement in the late 1980s and early 1990s. But this was no overnight hit, nor was it the brainchild of a single Silicon Valley entrepreneur.The ideas behind modern AI—neural networks and machine learning—have roots you can trace to the last stages of World War II. Back then, academics were beginning to build computing systems meant to store and process information in ways similar to the human brain. I worked on an automated system for reading checks with character recognition.Pomerleau demonstrates his self-driving car in 1995.METZ: At Carnegie Mellon, a guy named Dean Pomerleau built a self-driving car in the late 1980s using a neural network. LeCun used the technology in the 1990s to build a system that could recognize handwritten digits, which ended up being used commercially by banks.So through the late ’80s and on into the ’90s, there was this resurgence in neural networks and their practical applications, LeCun’s work being the prime example. My first encounter with Yoshua was when he published the same thing, or more or less the same thing, four years after one of my students published it. And then a couple of years later there was a showdown at a conference where all of this came out. Over the decades, the technology had its ups and downs, but it failed to capture the attention of computer scientists broadly until around 2012, thanks to a handful of stubborn researchers who weren’t afraid to look foolish. What you do in science is you clarify things. (Bengio has denied Schmidhuber’s claims.)LECUN: The problem back then was that the methods required complicated software, lots of data, and powerful computers. That was kind of a dark period for Geoff, Yoshua, and I. We were not bitter, but perhaps a little sad that people didn’t want to see what we all thought was an obvious advantage. The number of people getting neural networks to work better was quite small.The Canadian Institute for Advanced Research got people like us from all over the world to talk to each other much more. It gave us something of a critical mass.LECUN: There was this very small community of people who had this in the back of their minds, that eventually neural nets would come back to the fore. They remained convinced that neural nets would light up the world and alter humanity’s destiny.While these pioneers were scattered around the globe, there happened to be an unusually large concentration of neural net devotees in Canada. We got together and decided that we should strive to rekindle interest in our work.But we needed a safe space to have little workshops and meetings to really develop our ideas before publishing them. Geoff published one in Science.Face-recognition test images from Hinton’s 2006 article in Science.TRUDEAU: Learning that Canada had quietly built the foundations of modern AI during this most recent winter, when people had given up and moved on, is sort of a validation for me of something Canada’s always done well, which is support pure science. We give really smart people the capacity to do smart things that may or may not end up somewhere commercial or concrete.HINTON: In 2006 in Toronto, we developed this method of training networks with lots of layers, which was more efficient. We had a paper that same year in Science that was very influential and helped back up our claims, which got a lot of people interested again. In 2009 two of the students in my lab developed a way of doing speech recognition using these deep nets, and that worked better than what was already out there.It was only a little better, but the existing technology had already been around for 30 years with no advances. The fact that these deep nets could do even slightly better over a few months meant that it was obvious that within a few years’ time they were going to progress even further. Like just about everyone else, Li Deng believed in a different form of AI known as symbolic AI. In this approach, you basically had to build speech recognition systems one line at a time, coding in specific behavior, and this was really slow going.Hinton mentioned that his neural-net approach to speech recognition was showing real progress. That’s only partly through luck: The government-backed Canadian Institute for Advanced Research (Cifar) attracted a small group of academics to the country by funding neural net research when it was anything but fashionable. It could learn to recognize words by analyzing the patterns in databases of spoken words, and it was performing faster than the symbolic, line-by-line work. Deng didn’t necessarily believe Hinton, but invited him and eventually two of his collaborators to Microsoft to work on the technology. Speech recognition took a huge leap forward at Microsoft, and then Google as well in 2010.Then, at the end of 2012, Hinton and two of his students have a huge image recognition breakthrough where they blew away previous techniques. That’s when not just Microsoft and Google but the rest of the industry woke up to these ideas.The thing to remember is, these are very old ideas. You need the data to train on, and you need the computing power to execute that training.LECUN: Why did it take so long? It backed computer scientists such as Geoffrey Hinton and Yann LeCun at the University of Toronto, Yoshua Bengio at the University of Montreal, and the University of Alberta’s Richard Sutton, encouraging them to share ideas and stick to their beliefs. Now we’re in a bit of a race between people trying to develop the algorithms and people trying to develop faster and faster computers. You have to sort of plan for your AI algorithms to work with the computers that will be available in 5 years’ and 10 years’ time.The computer has to have a sense of what’s good and what’s bad, and so you give it a special signal called a reward. It’s where a purpose comes from.A neural net is where you store the learning, and reinforcement is how you decide what changes you’d like to make.BENGIO: We’re still a long way from the kind of unsupervised learning that Geoff, Yann, and I dream about. They came up with many of the concepts that fueled the AI revolution, and all are now considered godfathers of the technology. A 2-year-old has intuitive notions of physics, gravity, pressure, and so on, and her parents never need to tell her about Isaac Newton’s equations for force and gravity. We interact with the world, observe, and somehow build a mental model of how things will unfold in the future, if we do this or that.We’re moving into a new phase of research into unsupervised learning, which connects with the work on reinforcement. We’re not just observing the world, but we’re acting in the world and then using the effect of those actions to figure out how it works. We can predict the consequences of our actions, which means that we don’t need to actually do something bad to realize it’s bad.So, what I’m after is finding ways to train machines so that they can learn by observation, so they can build those kind of predictive models of the world. You could say that the ability to predict is really the essence of intelligence, combined with the ability to act on your predictions.LECUN: It’s quite possible we’re going to make some significant progress over the next 3 years, 5 years, 10 years, or 15 years—something fairly nearby.

Read More

Can AI Learn to Understand Emotions?

Can AI Learn to Understand Emotions?

Growing up in Egypt in the 1980s, Rana el Kaliouby was fascinated by hidden languages—the rapid-fire blinks of 1s and 0s computers use to transform electricity into commands and the infinitely more complicated nonverbal cues that teenagers use to transmit volumes of hormone-laden information to each other. Today, el Kaliouby is the CEO of Affectiva, a company that’s building the type of emotionally intelligent AI systems Picard envisioned two decades ago. Affectiva’s software measures a user’s emotional response through algorithms that identify key facial landmarks and analyze pixels in those regions to classify facial expressions. Combinations of those facial expressions are then mapped to any of seven different emotions as well as some complex cognitive states such as drowsiness and distraction. “I see that our emotional AI technology can be a core component of online learning systems, health wearables even,” el Kaliouby says. “Imagine if your Fitbit was smart about when it told you to go to sleep or when you needed to get snacks. It could say, ‘Oh, I see that today is going to be a really busy day for you and you’re going to be stressed. “If you take chess or Go—these games in AI that people think are so hard to solve—those are nothing to compared to what can happen in a few minutes in facial expressions,” says Rosalind Picard, founder and director of the Affective Computing Research Group at the MIT Media Lab. AI is more sophisticated today—last year Google’s AlphaZero algorithm taught itself the game and defeated a world champion chess program called Stockfish in just four hours—but analyzing metrics like facial expression in real time “isn’t even in the same league” Picard says. Each expression is created through a combination of more than 40 distinct muscle movements ranging from eyebrow furrowing to nose wrinkling to lip puckering. When she wasn’t allowed to date, el Kaliouby studied her peers the same way that she did the Atari. While our brains subconsciously process complex emotions and their intensities, teaching an artificial neural network to wade through that tsunami of data is an extraordinary technological challenge, one that’s further complicated by the fact that nonverbal communication varies between cultures. Despite the challenges, artificial emotional intelligence is a technological brass ring for a growing number of companies and researchers. While the field is in many ways still in its infancy, serious resources are being devoted to developing tools that can analyze and predict emotional response. These emerging tools include apps that forecast when students will be stressed out, vocal analysis software that helps diagnose mania and schizophrenia, and programs that predict suicide risk based on social media posts. “There’s just a huge, huge amount of data and research that has to happen before it’s going to be something that our computers are smart about,” Picard says. While el Kaliouby was fighting to be taken seriously as a computer scientist in Egypt, Rosalind “Roz” Picard was in Boston waging a somewhat similar war. Picard spent her early days at the MIT Media Lab building mathematical models that emulate how the brain detects patterns from data it collects from the outside world. “I was always the first one to say ‘Oh, he has a crush on her’ because of all of the gestures and the eye contact,” she says. This looks really important for AI and computer intelligence, and I sure don’t want to do it,’ ” Picard says. “This would totally destroy my career as a woman. She began to do it herself, testing ways to capture data on genuine, spontaneous emotions and applying the same machine learning techniques she had used in previous research. It featured actors making hundreds of different facial expressions—a sort of library originally compiled to teach children on the autism spectrum how to read nonverbal cues. Following in the footsteps of her parents, both computer scientists, el Kaliouby knew that her knowledge of programming languages would be a foundational skill for her career. El Kaliouby finished her Ph.D. and embarked on a three-year stint at the MIT Media Lab, flying between Egypt to Boston while creating the next iteration of MindReader. Picard, in the meantime, had already developed several new tools for capturing emotions in data computers could read, including a set of sweatbands embedded with sensors to measure skin conductance. Worn on the palm of the hand, the sensors picked up changes in electrical conductivity that happen when someone becomes psychologically aroused and begins to sweat. Believing that MindReader and the biometric sensors could be used to help children on the autism spectrum learn to navigate social situations and control their emotional responses, el Kaliouby and Picard began a multi-year study. They were overwhelmed by how many organizations in industries ranging from retail to banking to robotics were interested in real-time data on their target audience’s emotional states. Affectiva's software maps a person's face and uses a series of neural networks to judge their emotion. Nearly a decade later, neuroscientist Dr. Ned T. Sahin is using Affectiva software to fulfill Roz and Rana’s early dreams of using the technology to help people on the autism spectrum. Sahin is the founder of Brain Power, a company that makes wearable life coaching technologies for people with brain and cognitive challenges. Sahin’s team has developed a suite of Google Glass augmented reality applications, some of which are powered by Affectiva algorithms, and many of which were originally designed for children but have applications for wider audiences. Affectiva algorithms identifies the emotion and shows the user one augmented reality emoji representing that feeling and another that doesn’t. People should be able to decide whether and when to use the technology, understand how their data is being used, and maintain a level of privacy. Affectiva’s licensing agreement prohibits the software from being used in security or surveillance, and it requires partner organizations to obtain explicit consent from users before deployment. But as the field expands, potential for misuse ratchets up. Groups like the IEEE Standards Association have issued guidelines for affective design that include calls for explicit consent and data transparency policies. As she processed her shock and grief, she began thinking about the loss from a more scientific perspective—how do users change their social media behaviors when a major life event happens? She expected to see shifts in positive social media activity, but her data also revealed that some new moms were expressing negative emotions, too, and posting less often than they were during pregnancy. Suspecting that these might be indicators of postpartum depression, De Choudhury, then working at Microsoft Research, conducted a separate study that compared data from Facebook posts to interviews with mothers before and after their children were born. Since then, De Choudhury has used social media to identify mental illness risk, including psychosis symptoms among patients with schizophrenia, while other researchers have created algorithms that detect signs of anxiety, depression, and post-traumatic stress disorder. “Social data can be helpful to clinicians and psychiatrists as well as public health workers because it gives them a sense of where are the risks,” De Choudhury says. Chris Danforth, co-director of the University of Vermont’s Computational Story Lab, believes that conversations around when and how to deploy predictive mental health algorithms are especially important as opaque organizations like Facebook move further into the field. She left Affective in 2013 and has since concentrated on several health-minded projects, including work with MIT research scientist Akane Sano to build predictive models of mood, stress, and depression using data from wearable sensors. The goal is to create models that anticipate changes in mood and physical health and to help users make evidence-based decisions to stay happier and healthier, she says. El Kaliouby was captivated by the idea that feelings could be measured, analyzed, and used to design systems that can genuinely connect with people. Earlier this year, Empatica received FDA clearance for the Embrace smartwatch, a device that uses skin conductance and other metrics combined with AI to monitor for seizures. Since launching the software development kit in 2014, the company has licensed its software to organizations in healthcare, gaming, education, market research, and retail, to name a few. Last year, Affectiva also joined the Partnership on AI—a technology consortium developing ethics and education protocols for AI systems—and el Kaliouby is currently working with the World Economic Forum to design an ethics curriculum for schools. “I just have this deep conviction that we’re building a new paradigm of how we communicate with computers,” el Kaliouby says. “That’s been the driving factor of my work.

Read More

India Working On Unmanned Tanks, Vessels, Robotic Weaponry For Future Wars – BloombergQuint

India Working On Unmanned Tanks, Vessels, Robotic Weaponry For Future Wars – BloombergQuint

This (AI) is where the future is going to be. We need to prepare ourselves for the next generation warfare which will be more and more technology driven, more and more automated and robotised," he told PTI.Like many other world powers, India had also started work on the application of AI to boost the capabilities of its armed forces, Kumar said, adding that unmanned aerial vehicles, unmanned naval vessels, unmanned tanks and automatic robotic rifles as weapon systems will have an extensive use in future wars."We need to create capabilities for all these platforms," he said.Military sources said the project would also include production of a range of unmanned platforms for the three services.They said the forces were strongly pushing for extensive applications of AI in their operational preparedness on a par with leading global military powers.The sources said the application of AI in the surveillance of India's borders with China and Pakistan could significantly ease the pressure on armed forces personnel guarding the sensitive frontiers.

Read More

What is machine learning? Everything you need to know – ZDNet

What is machine learning? Everything you need to know – ZDNet

From driving cars to translating speech, machine learning is driving an explosion in the capabilities of artificial intelligence — helping software make sense of the messy and unpredictable real world. One of the most obvious demonstrations of the power of machine learning are virtual assistants, such as Apple's Siri, Amazon's Alexa, the Google Assistant, and Microsoft Cortana. These exploitations include: computer vision for driverless cars, drones and delivery robots; speech and language recognition and synthesis for chatbots and service robots; facial recognition for surveillance in countries like China; helping radiologists to pick out tumors in x-rays, aiding researchers in spotting genetic sequences related to diseases and identifying molecules that could lead to more effective drugs in healthcare; allowing for predictive maintenance on infrastructure by analyzing IoT sensor data; underpinning the computer vision that makes the cashierless Amazon Go supermarket possible, offering reasonably accurate transcription and translation of speech for business meetings — the list goes on and on. Are machine-learning systems objective? As you'd expect, the choice and breadth of data used to train systems will influence the tasks they are suited to. For example, in 2016 Rachael Tatman, a National Science Foundation Graduate Research Fellow in the Linguistics Department at the University of Washington, found that Google's speech-recognition system performed better for male voices than female ones when auto-captioning a sample of YouTube videos, a result she ascribed to 'unbalanced training sets' with a preponderance of male speakers. As machine-learning systems move into new areas, such as aiding medical diagnosis, the possibility of systems being skewed towards offering a better service or fairer treatment to particular groups of people will likely become more of a concern. Alongside machine learning, there are various other approaches used to build AI systems, including evolutionary computation, where algorithms undergo random mutations and combinations between generations in an attempt to "evolve" optimal solutions, and expert systems, where computers are programmed with rules that allow them to mimic the behavior of a human expert in a specific domain, for example an autopilot system flying a plane.What are the main types of machine learning? A heavily recommended course for beginners to teach themselves the fundamentals of machine learning is this free Stanford University and Coursera lecture series by AI expert and Google Brain founder Andrew Ng. Another highly-rated free online course, praised for both the breadth of its coverage and the quality of its teaching, is this EdX and Columbia University introduction to machine learning, although students do mention it requires a solid knowledge of math up to university level. All of the major cloud platforms — Amazon Web Services, Microsoft Azure and Google Cloud Platform — provide access to the hardware needed to train and run machine-learning models, with Google letting Cloud Platform users test out its Tensor Processing Units — custom chips whose design is optimized for training and running machine-learning models. This cloud-based infrastructure includes the data stores needed to hold the vast amounts of training data, services to prepare that data for analysis, and visualization tools to display the results clearly. Newer services even streamline the creation of custom machine-learning models, with Google recently revealing a service that automates the creation of AI models, called Cloud AutoML. There are a wide variety of software frameworks for getting started with training and running machine-learning models, typically for the programming languages Python, R, C++, Java and MATLAB. During training for supervised learning, systems are exposed to large amounts of labelled data, for example images of handwritten figures annotated to indicate which number they correspond to. Given sufficient examples, a supervised-learning system would learn to recognize the clusters of pixels and shapes associated with each number and eventually be able to recognize handwritten numbers, able to reliably distinguish between the numbers 9 and 4 or 6 and 8. However, training these systems typically requires huge amounts of labelled data, with some systems needing to be exposed to millions of examples to master a task.As a result, the datasets used to train these systems can be vast, with Google's Open Images Dataset having about nine million images, its labeled video repository YouTube-8M linking to seven million labeled videos and ImageNet, one of the early databases of this kind, having more than 14 million categorized images. The laborious process of labeling the datasets used in training is often carried out using crowdworking services, such as Amazon Mechanical Turk, which provides access to a large pool of low-cost labor spread across the globe. However, Facebook's approach of using publicly available data to train systems could provide an alternative way of training systems using billion-strong datasets without the overhead of manual labeling.What is unsupervised learning? The importance of huge sets of labelled data for training machine-learning systems may diminish over time, due to the rise of semi-supervised learning. The labelled data is used to partially train a machine-learning model, and then that partially trained model is used to label the unlabelled data, a process called pseudo-labelling. The viability of semi-supervised learning has been boosted recently by Generative Adversarial Networks ( GANs), machine-learning systems that can use labelled data to generate completely new data, for example creating new images of Pokemon from existing images, which in turn can be used to help train a machine-learning model. Were semi-supervised learning to become as effective as supervised learning, then access to huge amounts of computing power may end up being more important for successfully training machine-learning systems than access to large, labelled datasets.What is reinforcement learning? A way to understand reinforcement learning is to think about how someone might learn to play an old school computer game for the first time, when they aren't familiar with the rules or how to control the game. The system is fed pixels from each game and determines various information about the state of the game, such as the distance between objects on screen. It then considers how the state of the game and the actions it performs in game relate to the score it achieves. Over the process of many cycles of playing the game, eventually the system builds a model of which actions will maximize the score in which circumstance, for instance, in the case of the video game Breakout, where the paddle should be moved to in order to intercept the ball.How does supervised machine learning work? Before training begins, you first have to choose which data to gather and decide which features of the data are important. A hugely simplified example of what data features are is given in this explainer by Google, where a machine learning model is trained to recognize the difference between beer and wine, based on two features, the drinks' color and their alcoholic volume (ABV). Each drink is labelled as a beer or a wine, and then the relevant data is collected, using a spectrometer to measure their color and hydrometer to measure their alcohol content. Those predictions could be answering whether a piece of fruit in a photo is a banana or an apple, spotting people crossing the road in front of a self-driving car, whether the use of the word book in a sentence relates to a paperback or a hotel reservation, whether an email is spam, or recognizing speech accurately enough to generate captions for a YouTube video. An important point to note is that the data has to be balanced, in this instance to have a roughly equal number of examples of beer and wine. Each have strengths and weaknesses depending on the type of data, for example some are suited to handling images, some to text, and some to purely numerical data.How does supervised machine-learning training work? Basically, the training process involves the machine-learning model automatically tweaking how it functions until it can make accurate predictions from data, in the Google example, correctly labeling a drink as beer or wine when the model is given a drink's color and ABV. In the following example, the model is used to estimate how many ice creams will be sold based on the outside temperature. The key difference from traditional computer software is that a human developer hasn't written code that instructs the system how to tell the difference between the banana and the apple. To predict how many ice creams will be sold in future based on the outdoor temperature, you can draw a line that passes through the middle of all these points, similar to the illustration below. Image: Nick Heath / ZDNet Once this is done, ice cream sales can be predicted at any temperature by finding the point at which the line passes through a particular temperature and reading off the corresponding sales at that point. Bringing it back to training a machine-learning model, in this instance training a linear regression model would involve adjusting the vertical position and slope of the line until it lies in the middle of all of the points on the scatter graph. At each step of the training process, the vertical distance of each of these points from the line is measured. If a change in slope or position of the line results in the distance to these points increasing, then the slope or position of the line is changed in the opposite direction, and a new measurement is taken. In this way, via many tiny adjustments to the slope and the position of the line, the line will keep moving until it eventually settles in a position which is a good fit for the distribution of all these points, as seen in the video below. Once this training process is complete, the line can be used to make accurate predictions for how temperature will affect ice cream sales, and the machine-learning model can be said to have been trained. While training for more complex machine-learning models such as neural networks differs in several respects, it is similar in that it also uses a "gradient descent" approach, where the value of "weights" that modify input data are repeatedly tweaked until the output values produced by the model are as close as possible to what is desired. Once training of the model is complete, the model is evaluated using the remaining data that wasn't used during training, helping to gauge its real-world performance. Instead a machine-learning model has been taught how to reliably discriminate between the fruits by being trained on a large amount of data, in this instance likely a huge number of images labelled as containing a banana or an apple. An example might be altering the extent to which the "weights" are altered at each step in the training process.What are neural networks and how are they trained? These underlie much of machine learning, and while simple models like linear regression used can be used to make predictions based on a small number of data features, as in the Google example with beer and wine, neural networks are useful when dealing with large sets of data with many features. Neural networks, whose structure is loosely inspired by that of the brain, are interconnected layers of algorithms, called neurons, which feed data into each other, with the output of the preceding layer being the input of the subsequent layer. For instance, consider the example of using machine learning to recognize handwritten numbers between 0 and 9. The first layer in the neural network might measure the color of the individual pixels in the image, the second layer could spot shapes, such as lines and curves, the next layer might look for larger components of the written number — for example, the rounded loop at the base of the number 6. This carries on all the way through to the final layer, which will output the probability that a given handwritten figure is a number between 0 and 9. See more: Special report: How to implement AI and machine learning (free PDF) The network learns how to recognize each component of the numbers during the training process, by gradually tweaking the importance of data as it flows between the layers of the network. At the end of each training cycle the system will examine whether the neural network's final output is getting closer or further away from what is desired — for instance is the network getting better or worse at identifying a handwritten number 6. To close the gap between between the actual output and desired output, the system will then work backwards through the neural network, altering the weights attached to all of these links between layers, as well as an associated value called bias. Data, and lots of it, is the key to making machine learning possible.What is the difference between AI and machine learning? Eventually this process will settle on values for these weights and biases that will allow the network to reliably perform a given task, such as recognizing handwritten numbers, and the network can be said to have "learned" how to carry out a specific taskAn illustration of the structure of a neural network and how training works. A subset of machine learning is deep learning, where neural networks are expanded into sprawling networks with a huge number of layers that are trained using massive amounts of data. It is these deep neural networks that have fueled the current leap forward in the ability of computers to carry out task like speech recognition and computer vision. The design of neural networks is also evolving, with researchers recently devising a more efficient design for an effective type of deep neural network called long short-term memory or LSTM, allowing it to operate fast enough to be used in on-demand systems like Google Translate. This resurgence comes on the back of a series of breakthroughs, with deep learning setting new records for accuracy in areas such as speech and language recognition, and computer vision. What's made these successes possible are primarily two factors, one being the vast quantities of images, speech, video and text that is accessible to researchers looking to train machine-learning systems. But even more important is the availability of vast amounts of parallel-processing power, courtesy of modern graphics processing units (GPUs), which can be linked together into clusters to form machine-learning powerhouses. An example of one of these custom chips is Google's Tensor Processing Unit (TPU), the latest version of which accelerates the rate at which machine-learning models built using Google's TensorFlow software library can infer information from data, as well as the rate at which they can be trained. These chips are not just used to train models for Google DeepMind and Google Brain, but also the models that underpin Google Translate and the image recognition in Google Photo, as well as services that allow the public to build machine learning models using Google's TensorFlow Research Cloud. The second generation of these chips was unveiled at Google's I/O conference in May last year, with an array of these new TPUs able to train a Google machine-learning model used for translation in half the time it would take an array of the top-end GPUs, and the recently announced third-generation TPUs able to accelerate training and inference even further. At the birth of the field of AI in the 1950s, AI was defined as any machine capable of performing a task that would typically require human intelligence. Perhaps the most famous demonstration of the efficacy of machine-learning systems was the 2016 triumph of the Google DeepMind AlphaGo AI over a human grandmaster in Go, a feat that wasn't expected until 2026. Over the course of a game of Go, there are so many possible moves that searching through each of them in advance to identify the best play is too costly from a computational standpoint. Training the deep-learning networks needed can take a very long time, requiring vast amounts of data to be ingested and iterated over as the system gradually refines its model in order to achieve the best outcome.

Read More

Online learning company Pluralsight spikes in debut on Nasdaq

Online learning company Pluralsight spikes in debut on Nasdaq

Shares of online learning company Pluralsight began trading on the Nasdaq at an opening price of $20 per share Thursday, roughly 34 percent above the $15 price at which Pluralsight priced shares in its initial public offering. At the end of the first quarter the company had 14,830 business customers, and at the end of last year it had more than 695,000 end users. The company maintains an "army of expert authors" around the world to keep courses up to date and relevant, Skonnard said. The arrival of Pluralsight on public markets suggests that public investors are still eager to invest in smaller technology companies. Pluralsight first filed to go public on April 16. On May 7, the company said it estimated it would price shares between $10 and $12 each. Finally, on May 16, the company announced the $15 pricing, coming in above the high end of the most recent range. While anyone can pay to take Pluralsight's courses online, the company focuses on education for many employees inside companies.

Read More
1 2 3 16