Ronald Summers, Senior Investigator at the National Institutes of Health (NIH) shares how they are trying to improve patient care by increasing the accuracy of radiologic diagnosis with advanced computer techniques. His group is using deep learning and NVIDIA GPUs to assist physicians make a more accurate diagnosis by developing software that improves diagnosis, reduce the chance of errors, and help underserved patients that have limited access to advanced radiology services. "With deep learning and GPU acceleration we've had a substantial improvement in the performance of all these computer programs to the point where the programs are getting pretty close to performing as well as the average physician," says Summers. Watch more scientists and researchers share how accelerated computing is benefiting their work at http://nvda.ws/2dbscA7

Google's DeepMind is teaming up with the makers of the StarCraft video game to train its artificial intelligence systems. The AI systems "playing" the game will need to learn strategies similar to those that humans need in the real world, DeepMind said. Its ultimate aim is to develop artificial intelligence that could solve any problem. It has previously taught algorithms to play a range of Atari computer games. StarCraft II, made by developer Blizzard, is a real-time strategy game in which players control one of three warring factions - humans, the insect-like Zerg, or aliens known as the Protoss.

Though Intel is the leader in data centers microprocessor market, with more than 90% market share, it has more recently lagged in the market for coprocessors, where Nvidia has generated momentum. Whereas Intel is dominant in the microprocessor market at large, Nvidia has successfully identified this coprocessor niche where it has garnered significant market share. This is reflected in their recent growth rates. While Intel witnessed a mere 13% year over year growth in its revenue from the Data Center Group in Q3 2017, Nvidia data center revenues nearly tripled during the same period. Of course, Intel's data center revenues are a vast multiple of Nvidia's.

Calculus is not prerequisite to learn a lot of ML algorithms such as KNN, Naive Bayes, Decsion Trees, Random Forest, boosting etc and similar methods. However, if you wanna go the route for neural nets or deep learning, you need to figure out calculus because algorithms such as backpropagation uses them a lot. Also Gradient boosting and similar type of algorithms will require calculus knowledge.

The New York Times Magazine features a beautifully prepared and presented article on the recent dramatic improvements in Google Translate brought about by deep learning technologies developed in the Google Brain division of Alphabet. But this is not simply an article about a breakthrough innovation, it is a nuanced discussion of the history, problems, and approaches, both failed and (sometimes) successful, that lie behind some of the most useful AI-supported facilities we use today. "…in 1943 it was shown that arrangements of simple artificial neurons could carry out basic logical functions. They could also, at least in theory, learn the way we do. With life experience, depending on a particular person's trials and errors, the synaptic connections among pairs of neurons get stronger or weaker. An artificial neural network could do something similar, by gradually altering, on a guided trial-and-error basis, the numerical relationships among artificial neurons. It wouldn't need to be preprogrammed with fixed rules. It would, instead, rewire itself to reflect patterns in the data it absorbed…. If you wanted something that could adapt, you didn't want to begin with the indoctrination of the rules of chess. You wanted to begin with very basic abilities -- sensory perception and motor control -- in the hope that advanced skills would emerge organically. Humans don't learn to understand language by memorizing dictionaries and grammar books, so why should we possibly expect our computers to do so? Google Brain was the first major commercial institution to invest in the possibilities embodied by this way of thinking about A.I." This piece is required reading for anyone interested in understanding cognitive computing from the inside.

ARM servers are devalued partly because many applications don't work with the chips. But ARM has acquired Allinea Software with the hope of partially resolving the compatibility issue. Allinea provides software development, debugging, and porting tools, which should make it easier for people to write applications for ARM-based servers and supercomputers. The acquisition will "provide a channel to thousands of developers using supercomputers and give us better first-hand knowledge of the issues being addressed as software is ported to new ARM-based systems," Javier Orensanz, general manager of the development solutions group at ARM, said in a blog entry. The development tools will also be used for ARM chips in deep-learning systems, which require large-scale server deployments for analytics. ARM's competition will come from deep-learning and HPC software tools offered by Intel, Nvidia, Google, Microsoft, and more recently, AMD.

DeepMind, a research lab that was acquired by Google for £400 million, has become a well known entity in the field of artificial intelligence (AI) for building agents that can learn and master games such as arcade classic "Space Invaders" and the ancient Chinese board game of "Go". Over the last year, the five-year-old company, which employs approximately 250 people in London, has been branching out and applying its self-learning algorithms to fields such as healthcare and energy. On the latter, it's helped Google to slash the electricity bill in its data centres worldwide and it's now exploring how it can help the National Grid to predict demand. But Demis Hassabis, DeepMind's cofounder and CEO, announced on Sunday that the company isn't about to turn its back on the gaming field any time soon. In fact, Hassabis wrote on Twitter that DeepMind has been busy improving the AlphaGo [AG] agent that beat Lee SeDol, the world's best Go player, earlier this year.

When I attend customer engagement and industry events, I inevitably field lots of questions that are close to the heart of a data scientist. Many executives are confused by the concepts of machine learning, deep learning, memory-based learning, and artificial intelligence. They wonder about the differences in these technologies, how everything fits together, and what they need to pay attention to. They wonder whether they need all of it or just some of it, and what they need to do to get started. And, yes, I hear people ask whether the ultimate goal is to replace humans with computers.

You might go to a cutting-edge machine learning research conference like NIPS hoping to find some mathematical insight that will help you take your deep learning system's performance to the next level. Unfortunately, as Andrew Ng reiterated to a live crowd of 1,000 attendees this past Monday, there is no secret AI equation that will let you escape your machine learning woes. All you need is some rigor, and much of what Ng covered is his remarkable NIPS 2016 presentation titled "The Nuts and Bolts of Building Applications using Deep Learning" is not rocket science. Andrew Ng delivers a powerful message at NIPS 2016. Andrew Ng's lecture at NIPS 2016 in Barcelona was phenomenal -- truly one of the best presentations I have seen in a long time.

Thus, given a sequence of frames corresponding to an utterance, the model is required to produce, for each frame, a probability distribution over the alphabet. During the training phase, the softmax outputs are fed into a CTC cost function (more on this shortly) which uses the actual transcripts to (i) score the model's predictions, and (ii) generate an error signal quantifying the accuracy of the model's predictions. The overall goal is to train the model to increase the overall score of its predictions relative to the actual transcripts. Training Empirically, we have found that using stochastic gradient descent with momentum paired with gradient clipping leads to the best performing models. Deeper networks (seven layers or more) also tend to perform better in general.