We introduce dropout compaction, a novel method for training feed-forward neural networks which realizes the performance gains of training a large model with dropout regularization, yet extracts a compact neural network for run-time efficiency. In the proposed method, we introduce a sparsity-inducing prior on the per unit dropout retention probability so that the optimizer can effectively prune hidden units during training. By changing the prior hyperparameters, we can control the size of the resulting network. We performed a systematic comparison of dropout compaction and competing methods on several real-world speech recognition tasks and found that dropout compaction achieved comparable accuracy with fewer than 50% of the hidden units, translating to a 2.5x speedup in run-time.

The dynamic duo of deep learning and search is on the move. Google, in fact, is already pressing forward. In late 2015, the search giant announced it had added the deep-learning enabled RankBrain algorithm to give Hummingbird a new dimension. Google doesn't sit still for long, so of course they've been experimenting since then. Google's plan for deep learning is AI-based and leans on humanizing the algorithm, training it to recognize images and even speech patterns.

The objective of this course is to give you a wholistic understanding of machine learning, covering theory, application, and inner workings of supervised, unsupervised, and deep learning algorithms. In this series, we'll be covering linear regression, K Nearest Neighbors, Support Vector Machines (SVM), flat clustering, hierarchical clustering, and neural networks. For each major algorithm that we cover, we will discuss the high level intuitions of the algorithms and how they are logically meant to work. Next, we'll apply the algorithms in code using real world data sets along with a module, such as with Scikit-Learn. Finally, we'll be diving into the inner workings of each of the algorithms by recreating them in code, from scratch, ourselves, including all of the math involved.

Radiologists, technicians and doctors are hard to find in small towns, so entrepreneurs are using machine learning and artificial intelligence to bridge the gap BENGALURU: As a researcher at Xerox Research Centre in Bengaluru, healthcare researcher Geetha Manjunath always wondered why thermal images weren't used to detect breast cancer. "I thought it would be a non-invasive, non-radiation and non-intrusive method," she says, adding that some women are reluctant to have regular mammograms. She spent three years doing her research and analysis and in 2016, co-founded Niramai, an artificial intelligence-based healthcare startup which uses thermal imaging and artificial intelligence (AI) to screen for breast cancer. Niramai has signed up a couple of hospitals for its subscription model. More than a billion diagnostic tests are done in India every year in thousands of small labs and centres, many without the right equipment.

Chatbots, also called Conversational Agents or Dialog Systems, are a hot topic. Microsoft is making big bets on chatbots, and so are companies like Facebook (M), Apple (Siri), Google, WeChat, and Slack. There is a new wave of startups trying to change how consumers interact with services by building consumer apps like Operator or x.ai, bot platforms like Chatfuel, and bot libraries like Howdy's Botkit. Microsoft recently released their own bot developer framework. Many companies are hoping to develop bots to have natural conversations indistinguishable from human ones, and many are claiming to be using NLP and Deep Learning techniques to make this possible.

It easier than ever before to train a neural network. However, it is rarely the case that you can just take code from a tutorial and directly make it work for your application. Interestingly, many of the most important tweaks are barely discussed in the academic literature but at the same time critical to make your product work. Therefore I thought it would be helpful for other people who plan to use deep learning in their business to understand some of these tweaks and tricks. This post is based on my talk I gave on May 10 at the Berlin.AI meetup (the slides are here).

There is definitely an established machine learning ecosystem, or, perhaps more accurately, a small set of established machine learning ecosystems. For research it would seem that the undisputed champion of machine learning ecosystems is centered on Python and its many libraries which support the data preparation and subsequent machine learning process itself, whether it be via scikit-learn, one of the many deep learning libraries available, or home-spun and highly specialized tools for achieving the same goals. This says nothing of the great support tools that grow up around the edges of the ecosystem, some of which become polished and useful enough to carve out their own eventual niche. As those in industry would be the first to let me know, Python is not the only option. There are Java-based tools (Deeplearning4j, Weka), those integrated with Apache Spark and/or Hadoop (MLlib, Mahout), C solutions (TensorFlow is written in C, as are many others in the Python ecosystem), and even those for Clojure, F#, Rust, and a whole host of other languages, environments, and ecosystems.

Does the president pass the Turing Test? When I listen to his answers to basic questions and compare those answers to a real human's, it's plain to see that he's a computer--most likely, my research suggests, a Tandy 1000 EX purchased from a RadioShack in Secaucus, NJ sometime in December 1986. If this is the case, it explains a lot of his more mystifying decision-making procedures. The neurologist Robert A. Burton sees plenty of evidence that the president uses machine learning, making him a rudimentary artificial intelligence: "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 … 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."

A'new deep learning' method, DeepCpG, designed by researchers at EMBL-EBI, the Babraham Institute and the Sanger Institute helps scientists better understand the epigenome – the biochemical activity around the genome. Published in Genome Biology, DeepCpG leverages'deep neural networks', a multi-layered machine learning model inspired by the brain. Machine learning provides a valuable tool for research into health and disease. Deep learning is one of the most active fields in machine learning, which has led to recent advancements in computer image classification, text translation and speech recognition. But deep learning also has major potential in computational biology, particularly for regulatory genomics and cellular imaging. "We now have this amazing'book' of the human genome, thanks to projects like 1000 Genomes, divided up nicely into chapters and annotated in parts.

AlphaGo is at it again. Google DeepMind's Go-playing AI has defeated Ke Jie, the world's number one player, in the first of three games played in Wuzhen, China. The AI won by just half a point – the smallest possible margin of victory – in a match that lasted four hours and fifteen minutes. Though the scoreline looks close, AlphaGo was in the lead from relatively early on in the game. Since the AI favours moves that are more likely to guarantee victory, it doesn't usually trounce its opponents. In March last year, AlphaGo beat Lee Sedol, one of the world's top Go players, winning four out of five matches.