Andrew Ng is one of the foremost thinkers on the topic of artificial intelligence. He founded and led the "Google Brain" project which developed massive-scale deep learning algorithms. In 2011, he led the development of Stanford University's main Massive Open Online Course (MOOC) platform. His course on Machine Learning would eventually reach an "enrollment" of over 100,000 students. That experience led Ng to co-found Coursera, a MOOC that partners with some of the top universities in the world to offer high quality online courses. Today, Coursera is the largest MOOC platform in the world.

Self-driving cars, house appliances that send notifications to your mobile devices, computer programs that can beat humans at computer games – you hear about such stuff in every artificial intelligence report and article you read, we know that. And until some machine can make you pancakes in the morning and wash the dishes for you, there is no reason to get that excited yet, is there? Can that machine clean your kitchen for you and make you spaghetti without burning the meatballs? That will be the day when that can happen! Artificial intelligence has tremendous potential, but it's still on the road and far from its destination of value-adding and easily doable automation of sophisticated computer based processes.

Blizzard's partnership with DeepMind, a firm that specializes in machine learning, has culminated in the release of the StarCraft II API, which is available now. This mechanism for enabling outside creators into integrate their apps into StarCraft II is going to open up the competitive sci-fi strategy game to researchers working in the field of A.I. At the 2016 BlizzCon fan gathering in Anaheim, Blizzard revealed that it was working with DeepMind to test the same sort of learning algorithms that helped the company's AlphaGo A.I. beat some of the top players of the complicated board game, Go. "On behalf of Blizzard Entertainment, the StarCraft II development team is very pleased to announce the release of the StarCraft II API," reads a Blizzard blog post. "We recognize the efforts made by researchers over the years to advance AI using the original StarCraft. With the StarCraft II API, we're providing powerful tools for researchers, gamers, and hobbyists to utilize the game as a platform to further advance the state of AI research."

This week I'm going to focus on how deep learning is used in self-driving cars. There are plenty of machine learning applications within that field, but I'm going to zoom in on one very cool technology: virtual testing. Let me cut to the chase: below's a video of my fully-autonomous car driving around in a virtual testing environment. Granted, while it does looks like a 1990s retro (pretty dull) driving game, the car is being controlled by a deep convolutional neural net (CNN) based on a modification a deep learning algorithm made by Nvidia. Let's dive in and take a look at what's going on here.

Probabilistic inversion within a multiple-point statistics framework is still computationally prohibitive for large-scale problems. To partly address this, we introduce and evaluate a new training-image based simulation and inversion approach for complex geologic media. Our approach relies on a deep neural network of the spatial generative adversarial network (SGAN) type. After training using a training image (TI), our proposed SGAN can quickly generate 2D and 3D unconditional realizations. A key feature of our SGAN is that it defines a (very) low-dimensional parameterization, thereby allowing for efficient probabilistic (or deterministic) inversion using state-of-the-art Markov chain Monte Carlo (MCMC) methods. A series of 2D and 3D categorical TIs is first used to analyze the performance of our SGAN for unconditional simulation. The speed at which realizations are generated makes it especially useful for simulating over large grids and/or from a complex multi-categorical TI. Subsequently, synthetic inversion case studies involving 2D steady-state flow and 3D transient hydraulic tomography are used to illustrate the effectiveness of our proposed SGAN-based probabilistic inversion. For the 2D case, the inversion rapidly explores the posterior model distribution. For the 3D case, the inversion recovers model realizations that fit the data close to the target level and visually resemble the true model well. Future work will focus on the inclusion of direct conditioning data and application to continuous TIs.

We present a novel regularization scheme for training deep neural networks. The parameters of neural networks are usually unconstrained and have a dynamic range dispersed over the real line. Our key idea is to control the expressive power of the network by dynamically quantizing the range and set of values that the parameters can take. We formulate this idea using a novel end-to-end approach that regularizes the traditional classification loss function. Our regularizer is inspired by the Minimum Description Length principle. For each layer of the network, our approach optimizes a translation and scaling factor along with integer-valued parameters. We empirically compare BitNet to an equivalent unregularized model on the MNIST and CIFAR-10 datasets. We show that BitNet converges faster to a superior quality solution. Additionally, the resulting model is significantly smaller in size due to the use of integer parameters instead of floats.

The predictive power of neural networks often costs model interpretability. Several techniques have been developed for explaining model outputs in terms of input features; however, it is difficult to translate such interpretations into actionable insight. Here, we propose a framework to analyze predictions in terms of the model's internal features by inspecting information flow through the network. Given a trained network and a test image, we select neurons by two metrics, both measured over a set of images created by perturbations to the input image: (1) magnitude of the correlation between the neuron activation and the network output and (2) precision of the neuron activation. We show that the former metric selects neurons that exert large influence over the network output while the latter metric selects neurons that activate on generalizable features. By comparing the sets of neurons selected by these two metrics, our framework suggests a way to investigate the internal attention mechanisms of convolutional neural networks.

This survey explores Procedural Content Generation via Machine Learning (PCGML), defined as the generation of game content using machine learning models trained on existing content. As the importance of PCG for game development increases, researchers explore new avenues for generating high-quality content with or without human involvement; this paper addresses the relatively new paradigm of using machine learning (in contrast with search-based, solver-based, and constructive methods). We focus on what is most often considered functional game content such as platformer levels, game maps, interactive fiction stories, and cards in collectible card games, as opposed to cosmetic content such as sprites and sound effects. In addition to using PCG for autonomous generation, co-creativity, mixed-initiative design, and compression, PCGML is suited for repair, critique, and content analysis because of its focus on modeling existing content. We discuss various data sources and representations that affect the resulting generated content. Multiple PCGML methods are covered, including neural networks, long short-term memory (LSTM) networks, autoencoders, and deep convolutional networks; Markov models, $n$-grams, and multi-dimensional Markov chains; clustering; and matrix factorization. Finally, we discuss open problems in the application of PCGML, including learning from small datasets, lack of training data, multi-layered learning, style-transfer, parameter tuning, and PCG as a game mechanic.

This post is part of an insideHPC guide that explores how to successfully launch a machine learning startup. The complete report, available here, covers how to get started, how to choose a framework, how to decide what applications and technology to use, and more. While artificial intelligence (AI), machine learning and deep learning are often thought of as being interchangeable, they do in fact relate to very different concepts. It all began in the 1950s with AI and the idea that a computer could be made to simulate human learning and intelligence. A subclass of that is machine learning, whereby a computer can take large amounts of data and use it begin to recognize patterns, make predictions on new data, and essentially'learn' for itself.

She joined as senior reporter in April 2014 having previously worked as assistant editor at Government Computing. This year has seen an artificial intelligence system beat professional poker players at a notoriously difficult game for machines to master. However this is just the latest round in the ongoing battle of human versus machine. In 2016 a system built by Google-owned AI company DeepMind called'AlphaGo' beat South Korean champion Lee Sedol at the fiendishly complex game'Go'. Sedol won just one game to AlphaGo's four across a five-match series.