Modeling agent behavior is central to understanding the emergence of complex phenomena in multiagent systems. Prior work in agent modeling has largely been task-specific and driven by hand-engineering domain-specific prior knowledge. We propose a general learning framework for modeling agent behavior in any multiagent system using only a handful of interaction data. Our framework casts agent modeling as a representation learning problem. Consequently, we construct a novel objective inspired by imitation learning and agent identification and design an algorithm for unsupervised learning of representations of agent policies. We demonstrate empirically the utility of the proposed framework in (i) a challenging high-dimensional competitive environment for continuous control and (ii) a cooperative environment for communication, on supervised predictive tasks, unsupervised clustering, and policy optimization using deep reinforcement learning.

The dominant, state-of-the-art collaborative filtering (CF) methods today mainly comprises neural models. In these models, deep neural networks, e.g.., multi-layered perceptrons (MLP), are often used to model nonlinear relationships between user and item representations. As opposed to shallow models (e.g., factorization-based models), deep models generally provide a greater extent of expressiveness, albeit at the expense of impaired/restricted information flow. Consequently, the performance of most neural CF models plateaus at 3-4 layers, with performance stagnating or even degrading when increasing the model depth. As such, the question of how to train really deep networks in the context of CF remains unclear. To this end, this paper proposes a new technique that enables training neural CF models all the way up to 20 layers and beyond. Our proposed approach utilizes a new hierarchical self-attention mechanism that learns introspective intra-feature similarity across all the hidden layers of a standard MLP model. All in all, our proposed architecture, SA-NCF (Self-Attentive Neural Collaborative Filtering) is a densely connected self-matching model that can be trained up to 24 layers without plateau-ing, achieving wide performance margins against its competitors. On several popular benchmark datasets, our proposed architecture achieves up to an absolute improvement of 23%-58% and 1.3x to 2.8x fold improvement in terms of nDCG@10 and Hit Ratio (HR@10) scores over several strong neural CF baselines.

Value Iteration Networks (VINs) are effective differentiable path planning modules that can be used by agents to perform navigation while still maintaining end-to-end differentiability of the entire architecture. Despite their effectiveness, they suffer from several disadvantages including training instability, random seed sensitivity, and other optimization problems. In this work, we reframe VINs as recurrent-convolutional networks which demonstrates that VINs couple recurrent convolutions with an unconventional max-pooling activation. From this perspective, we argue that standard gated recurrent update equations could potentially alleviate the optimization issues plaguing VIN. The resulting architecture, which we call the Gated Path Planning Network, is shown to empirically outperform VIN on a variety of metrics such as learning speed, hyperparameter sensitivity, iteration count, and even generalization. Furthermore, we show that this performance gap is consistent across different maze transition types, maze sizes and even show success on a challenging 3D environment, where the planner is only provided with first-person RGB images.

Q: What is Deep Vision Data? A: Deep Vision Data creates synthetic training data for machine learning systems such as neural networks. Neural networks are computer systems that aren't explicitly programmed, instead they are "trained" by providing them thousands or millions of examples. Self-driving cars, personal assistants like Alexa, Siri and Google Home, language translation and facial recognition are all applications of a class of machine learning called deep neural networks. Q: What is synthetic training data?

Artificial intelligence (AI) is best known for its ability to see (as in driverless cars) and listen (as in Alexa and other home assistants). My colleagues and I are developing an AI system that can smell human breath and learn how to identify a range of illness-revealing substances that we might breathe out. The sense of smell is used by animals and even plants to identify hundreds of different substances that float in the air. But compared to that of other animals, the human sense of smell is far less developed and certainly not used to carry out daily activities. For this reason, humans aren't particularly aware of the richness of information that can be transmitted through the air, and can be perceived by a highly sensitive olfactory system.

A popular notion about machine learning models is the interpretability -- statistical models like logistic regression yield interpretable models. Historically, financial sector which relies heavily on interpretability uses machine learning models because of its ability to offer an audit trail. On the other hand, neural networks are dubbed as the black boxes since there is no real understanding of how the output was achieved. In other words, one may not be able to ascertain how the exact model works inside and out, but know the learning algorithm that created it. However, according to Zachary Chase Lipton, a Ph.D student at UCSD, machine learning algorithms, for example decision trees, often championed for their interpretability, can also be similarly opaque.

The power of modern search engines is undeniable: you can summon knowledge from the internet at a moment's notice. There are many situations where search is relegated to strict keyword search, or when the objects aren't text, search may not be available. Furthermore, strict keyword search doesn't allow the user to search semantically, which means information is not as discoverable. Today, we share a reproducible, minimally viable product that illustrates how you can enable semantic search for arbitrary objects! Concretely, we will show you how to create a system that searches python code semantically -- but this approach can be generalized to other entities (such as pictures or sound clips).

While standard machine learning can parse data and extract information to make decisions, it must be guided to know when a prediction it makes is correct or incorrect. Deep learning, on the other hand, can learn not only to make predictions, but also whether those predictions are likely to be accurate, and adjust the way it interprets to improve its predictions.

DeepMind's foray into digital health services continues to raise concerns. The latest worries are voiced by a panel of external reviewers appointed by the Google-owned AI company to report on its operations after its initial data-sharing arrangements with the U.K.'s National Health Service (NHS) ran into a major public controversy in 2016. The DeepMind Health Independent Reviewers' 2018 report flags a series of risks and concerns, as they see it, including the potential for DeepMind Health to be able to "exert excessive monopoly power" as a result of the data access and streaming infrastructure that's bundled with provision of the Streams app -- and which, contractually, positions DeepMind as the access-controlling intermediary between the structured health data and any other third parties that might, in the future, want to offer their own digital assistance solutions to the Trust. While the underlying FHIR (aka, fast healthcare interoperability resource) deployed by DeepMind for Streams uses an open API, the contract between the company and the Royal Free Trust funnels connections via DeepMind's own servers, and prohibits connections to other FHIR servers. A commercial structure that seemingly works against the openness and interoperability DeepMind's co-founder Mustafa Suleyman has claimed to support.

Artificial intelligence (AI) is in the midst of an undeniable surge in popularity, and enterprises are becoming particularly interested in a form of AI known as deep learning. According to Gartner, AI will likely generate $1.2 trillion in business value for enterprises in 2018, 70 percent more than last year. "AI promises to be the most disruptive class of technologies during the next 10 years due to advances in computational power, volume, velocity and variety of data, as well as advances in deep neural networks (DNNs)," said John-David Lovelock, research vice president at Gartner. Those deep neural networks are used for deep learning, which most enterprises believe will be important for their organizations. A 2018 O'Reilly report titled How Companies Are Putting AI to Work through Deep Learning found that only 28 percent of enterprises surveyed were already using deep learning.