We propose SWA-Gaussian (SWAG), a simple, scalable, and general purpose approach for uncertainty representation and calibration in deep learning. Stochastic Weight Averaging (SWA), which computes the first moment of stochastic gradient descent (SGD) iterates with a modified learning rate schedule, has recently been shown to improve generalization in deep learning. With SWAG, we fit a Gaussian using the SWA solution as the first moment and a low rank plus diagonal covariance also derived from the SGD iterates, forming an approximate posterior distribution over neural network weights; we then sample from this Gaussian distribution to perform Bayesian model averaging. We empirically find that SWAG approximates the shape of the true posterior, in accordance with results describing the stationary distribution of SGD iterates. Moreover, we demonstrate that SWAG performs well on a wide variety of computer vision tasks, including out of sample detection, calibration, and transfer learning, in comparison to many popular alternatives including MC dropout, KFAC Laplace, and temperature scaling.

A central challenge in reinforcement learning is discovering effective policies for tasks where rewards are sparsely distributed. We postulate that in the absence of useful reward signals, an effective exploration strategy should seek out {\it decision states}. These states lie at critical junctions in the state space from where the agent can transition to new, potentially unexplored regions. We propose to learn about decision states from prior experience. By training a goal-conditioned policy with an information bottleneck, we can identify decision states by examining where the model actually leverages the goal state. We find that this simple mechanism effectively identifies decision states, even in partially observed settings. In effect, the model learns the sensory cues that correlate with potential subgoals. In new environments, this model can then identify novel subgoals for further exploration, guiding the agent through a sequence of potential decision states and through new regions of the state space.

Learning the generative model, i.e., the underlying data generating distribution, based on large amounts of data is one the fundamental task in machine learning and statistics [46].Recent advances in deep generative models have provided novel techniques for unsupervised and semi-supervised learning, with broad application varying from image synthesis [44], semantic image editing [60], image-to-image translation [61] to low-level image processing [29]. Implicit deep generative model is a powerful and flexible framework to approximate the target distribution by learning deep samplers [38] including Generative adversarialnetworks (GAN) [16] and likelihood based models, such as variational auto-encoders (VAE) [23] and flow based methods [11], as their main representatives. The above mentioned implicit deep generative models focus on learning a deterministic or stochastic nonlinear mapping that can transform low dimensional latent samples from referenced simple distribution to samples that closely match the target distribution. GANs build a minmax two player game between the generator and discriminator. During the training, the generator transforms samples from a simple reference distribution into samples that would hopefully to deceive the discriminator, while the discriminator conducts a differential two-sample test to distinguish the generated samples from the observed samples. The objective of vanilla GANs amounts to the Jensen-Shannon (JS) divergence between the learned distribution and target distributions. The vanilla GAN generates sharp image samples but suffers form the instability issues [3]. A myriad of extensions to vanilla GANs have been investigated, both theoretically or empirically, in order to achieve a stable training and high quality sample generation.

Recent graph neural networks implement convolutional layers based on polynomial filters operating in the spectral domain. In this paper, we propose a novel graph convolutional layer based on auto-regressive moving average (ARMA) filters that, compared to the polynomial ones, provides a more flexible response thanks to a rich transfer function that accounts for the concept of state. We implement the ARMA filter with a recursive and distributed formulation, obtaining a convolutional layer that is efficient to train, is localized in the node space and can be applied to graphs with different topologies. In order to learn more abstract and compressed representations in deeper layers of the network, we alternate pooling operations based on node decimation to reduce the dimensionality of the node space with convolutions on coarsened versions of the original graph. We consider three major graph inference problems: semi-supervised node classification, graph classification, and graph signal classification. Results show that the proposed graph neural network with ARMA filters outperform those based on polynomial filters and sets the new state-of-the-art in several tasks.

We consider a novel approach to high-level robot task execution for a robot assistive task. In this work we explore the problem of learning to predict the next subtask by introducing a deep model for both sequencing goals and for visually evaluating the state of a task. We show that deep learning for monitoring robot tasks execution very well supports the interconnection between task-level planning and robot operations. These solutions can also cope with the natural non-determinism of the execution monitor. We show that a deep execution monitor leverages robot performance. We measure the improvement taking into account some robot helping tasks performed at a warehouse.

Can multilayer neural networks -- typically constructed as highly complex structures with many nonlinearly activated neurons across layers -- behave in a non-trivial way that yet simplifies away a major part of their complexities? In this work, we uncover a phenomenon in which the behavior of these complex networks -- under suitable scalings and stochastic gradient descent dynamics -- becomes independent of the number of neurons as this number grows sufficiently large. We develop a formalism in which this many-neurons limiting behavior is captured by a set of equations, thereby exposing a previously unknown operating regime of these networks. While the current pursuit is mathematically non-rigorous, it is complemented with several experiments that validate the existence of this behavior.

Motivation: Sepsis is a life-threatening host response to infection associated with high mortality, morbidity and health costs. Its management is highly time-sensitive since each hour of delayed treatment increases mortality due to irreversible organ damage. Meanwhile, despite decades of clinical research robust biomarkers for sepsis are missing. Therefore, detecting sepsis early by utilizing the affluence of high-resolution intensive care records has become a challenging machine learning problem. Recent advances in deep learning and data mining promise a powerful set of tools to efficiently address this task. Results: This paper proposes two approaches for the early detection of sepsis: a new deep learning model (MGP-TCN) and a data mining model (DTW-KNN). MGP-TCN employs a temporal convolutional network as embedded in a Multitask Gaussian Process Adapter framework, making it directly applicable to irregularly spaced time series data. Our DTW-KNN is an ensemble approach that employs dynamic time warping. We then frame the timely detection of sepsis as a supervised time series classification task. For this, we derive the most recent sepsis definition in an hourly resolution to provide the first fully accessible early sepsis detection environment. Seven hours before sepsis onset, our methods MGP-TCN/DTW-KNN improve area under the precision--recall curve from 0.25 to 0.35/0.40 over the state of the art. This demonstrates that they are well-suited for detecting sepsis in the crucial earlier stages when management is most effective.

ETH researchers have developed a new deep learning benchmarking environment – Deep500 – they say is "the first distributed and reproducible benchmarking system for deep learning, [and] provides software infrastructure to utilize the most powerful supercomputers for extreme-scale workloads." The researchers used CSCS Piz Daint supercomputer in developing the benchmark, have made the code freely available on GitHub, and last week published a detailed analysis of their approach (A Modular Benchmarking Infrastructure for High-Performance and Reproducible Deep Learning)[i]. "Deep500 [is] the first customizable bench- marking infrastructure that enables fair comparison of the plethora of deep learning frameworks, algorithms, libraries, and techniques," write the researchers. "The key idea behind Deep500 is its modular design, where deep learning is factorized into four distinct levels: operators, network processing, training, and distributed training. Our evaluation illustrates that Deep500 is customizable (enables combining and benchmarking different deep learning codes) and fair (uses carefully selected metrics). Moreover, Deep500 is fast (incurs negligible overheads), verifiable (offers infrastructure to analyze correctness), and reproducible."

If you've ever played the card game Hanabi, you'll understand when I say it's unlike any other. It's a collaborative game in which you have full view of everyone else's hands but not your own. To win the game, each player must give the others hints about their hands over a limited number of rounds to arrange all the cards in a specific order. That's why researchers at Google Brain and DeepMind think it's the perfect game for AI to tackle next. In a new paper, they argue that unlike the other games AI has mastered, such as chess, Go, and poker, Hanabi requires theory of mind and a higher level of reasoning.

"Generative A.I. systems are making a wide range of creative skills more accessible. Creative.AI co-founder Samim A Winiger gives a quick introduction to creativity in relation to machine learning and AI. Today it is possible to generate virtual worlds with billions of unique planets. But how do we think about creativity and how do we build systems for it? One example Samim gives is a music mood agent he worked on for a contemporary band.