There is a previously identified equivalence between wide fully connected neural networks (FCNs) and Gaussian processes (GPs). This equivalence enables, for instance, test set predictions that would have resulted from a fully Bayesian, infinitely wide trained FCN to be computed without ever instantiating the FCN, but by instead evaluating the corresponding GP. In this work, we derive an analogous equivalence for multi-layer convolutional neural networks (CNNs) both with and without pooling layers, and achieve state of the art results on CIFAR10 for GPs without trainable kernels. We also introduce a Monte Carlo method to estimate the GP corresponding to a given neural network architecture, even in cases where the analytic form has too many terms to be computationally feasible. Surprisingly, in the absence of pooling layers, the GPs corresponding to CNNs with and without weight sharing are identical. As a consequence, translation equivariance in finite-channel CNNs trained with stochastic gradient descent (SGD) has no corresponding property in the Bayesian treatment of the infinite channel limit - a qualitative difference between the two regimes that is not present in the FCN case. We confirm experimentally, that while in some scenarios the performance of SGD-trained finite CNNs approaches that of the corresponding GPs as the channel count increases, with careful tuning SGD-trained CNNs can significantly outperform their corresponding GPs, suggesting advantages from SGD training compared to fully Bayesian parameter estimation.

Past work in relation extraction mostly focuses on binary relation between entity pairs within single sentence. Recently, the NLP community has gained interest in relation extraction in entity pairs spanning multiple sentences. In this paper, we propose a novel architecture for this task: inter-sentential dependency-based neural networks (iDepNN). iDepNN models the shortest and augmented dependency paths via recurrent and recursive neural networks to extract relationships within (intra-) and across (inter-) sentence boundaries. Compared to SVM and neural network baselines, iDepNN is more robust to false positives in relationships spanning sentences. We evaluate our models on four datasets from newswire (MUC6) and medical (BioNLP shared task) domains that achieve state-of-the-art performance and show a better balance in precision and recall for inter-sentential relationships. We perform better than 11 teams participating in the BioNLP shared task 2016 and achieve a gain of 5.2% (0.587 vs 0.558) in F1 over the winning team. We also release the crosssentence annotations for MUC6.

Humans are experts at high-fidelity imitation -- closely mimicking a demonstration, often in one attempt. Humans use this ability to quickly solve a task instance, and to bootstrap learning of new tasks. Achieving these abilities in autonomous agents is an open problem. In this paper, we introduce an off-policy RL algorithm (MetaMimic) to narrow this gap. MetaMimic can learn both (i) policies for high-fidelity one-shot imitation of diverse novel skills, and (ii) policies that enable the agent to solve tasks more efficiently than the demonstrators. MetaMimic relies on the principle of storing all experiences in a memory and replaying these to learn massive deep neural network policies by off-policy RL. This paper introduces, to the best of our knowledge, the largest existing neural networks for deep RL and shows that larger networks with normalization are needed to achieve one-shot high-fidelity imitation on a challenging manipulation task. The results also show that both types of policy can be learned from vision, in spite of the task rewards being sparse, and without access to demonstrator actions.

This paper is about alerting acoustic event detection and sound source localisation in an urban scenario. Specifically, we are interested in spotting the presence of horns, and sirens of emergency vehicles. In order to obtain a reliable system able to operate robustly despite the presence of traffic noise, which can be copious, unstructured and unpredictable, we propose to treat the spectrograms of incoming stereo signals as images, and apply semantic segmentation, based on a Unet architecture, to extract the target sound from the background noise. In a multi-task learning scheme, together with signal denoising, we perform acoustic event classification to identify the nature of the alerting sound. Lastly, we use the denoised signals to localise the acoustic source on the horizon plane, by regressing the direction of arrival of the sound through a CNN architecture. Our experimental evaluation shows an average classification rate of 94%, and a median absolute error on the localisation of 7.5{\deg} when operating on audio frames of 0.5s, and of 2.5{\deg} when operating on frames of 2.5s. The system offers excellent performance in particularly challenging scenarios, where the noise level is remarkably high.

Information theory has been very successful in obtaining performance limits for various problems such as communication, compression and hypothesis testing. Likewise, stochastic control theory provides a characterization of optimal policies for Partially Observable Markov Decision Processes (POMDPs) using dynamic programming. However, finding optimal policies for these problems is computationally hard in general and thus, heuristic solutions are employed in practice. Deep learning can be used as a tool for designing better heuristics in such problems. In this paper, the problem of active sequential hypothesis testing is considered. The goal is to design a policy that can reliably infer the true hypothesis using as few samples as possible by adaptively selecting appropriate queries. This problem can be modeled as a POMDP and bounds on its value function exist in literature. However, optimal policies have not been identified and various heuristics are used. In this paper, two new heuristics are proposed: one based on deep reinforcement learning and another based on a KL-divergence zero-sum game. These heuristics are compared with state-of-the-art solutions and it is demonstrated using numerical experiments that the proposed heuristics can achieve significantly better performance than existing methods in some scenarios.

ARTIFICIAL intelligence breakthroughs have become a regular occurrence in recent years. One of the most impressive achievements so far was in 2016, when Google DeepMind's AlphaGo AI beat champion Lee Sedol at one of the world's most complex games, Go. The feat made headlines around the world as an example of machines besting humans, but in some sense it wasn't a fair fight.

No matter the industry, data science has become a universal toolkit for businesses. Data analytics and machine learning give organizations insights and answers that shape their day-to-day actions and future plans. Being data-driven has become essential to lead any industry. While the world's data doubles each year, CPU computing has hit a brick wall with the end of Moore's law. For this reason, scientific computing and deep learning have turned to NVIDIA GPU acceleration.

This spring I presented a talk entitled "Effective Transfer Learning for NLP" at ODSC East. The talk was intended to demonstrate how surprisingly effective pre-trained word and document embeddings are at low training data volumes, and to lay out a set of practical recommendations for applying these techniques to your own tasks. Thanks to some excellent research by Alec Radford and the team at OpenAI, our recommendations are beginning to change. To explain why the tides are shifting, let's first walk through the rubric we use at Indico to evaluate whether or not a novel machine learning method is viable for industry use. Let's see how well pre-trained word document embeddings satisfy these requirements: In short, using pre-trained embeddings is computationally cheap and performs well at the lower extremes of training data availability, but using static representations imposes an unfortunate cap on the benefit gained from additional training data.

Most of the games that machines can now challenge humans in are strategic, but slow: Chess, Go and poker, unless played in very specific settings, have no time constraints on player moves. That is what has made the work of research group OpenAI, in online team brawler Dota 2 - which requires real-time decision-making between potentially dozens of choices in a single frame - so different. OpenAI's bots, the OpenAI Five, went head-to-head against teams of professional players at Dota 2's annual championship, The International, this August. Although the bots lost, the matches provided an insight into how reinforcement learning is changing the game when it comes to artificial intelligence. It's safe to say that AI has a reputation in gaming: many players consider a match to be an instant loss if they have to play with a bot, and a disconnect is often accompanied by "GG".

Early I brought up Howard Gardner's theory of multiple intelligences. That is, humans exhibit strengths in different kinds of intelligences. Specifically these are interpersonal, intrapersonal, verbal, logical, spatial, rhythmic, naturalistic and kinaesthetic intelligence. Clearly there are many kinds of ways of thinking, each with their own strengths. Therefore, one may ask if we can use this notion of multiple intelligences to explore the different ways that AGI research may evolve.