Computing equilibrium states for many-body systems, such as molecules, is a long-standing challenge. In the absence of methods for generating statistically independent samples, great computational effort is invested in simulating these systems using, for example, Markov chain Monte Carlo. We present a probabilistic model that generates such samples for molecules from their graph representations. Our model learns a low-dimensional manifold that preserves the geometry of local atomic neighborhoods through a principled learning representation that is based on Euclidean distance geometry. We create a new dataset for molecular conformation generation with which we show experimentally that our generative model achieves state-of-the-art accuracy. Finally, we show how to use our model as a proposal distribution in an importance sampling scheme to compute molecular properties.

Understanding the asymptotic behavior of wide networks is of considerable interest. In this work, we present a general method for analyzing this large width behavior. The method is an adaptation of Feynman diagrams, a standard tool for computing multivariate Gaussian integrals. We apply our method to study training dynamics, improving existing bounds and deriving new results on wide network evolution during stochastic gradient descent. Going beyond the strict large width limit, we present closed-form expressions for higher-order terms governing wide network training, and test these predictions empirically.

We present simple reconstruction networks for multi-coil data by extending deep cascade of CNN's and exploiting the data consistency layer. In particular, we propose two variants, where one is inspired by POCSENSE and the other is calibration-less. We show that the proposed approaches are competitive relative to the state of the art both quantitatively and qualitatively.

Graph node embedding aims at learning a vector representation for all nodes given a graph. It is a central problem in many machine learning tasks (e.g., node classification, recommendation, community detection). The key problem in graph node embedding lies in how to define the dependence to neighbors. Existing approaches specify (either explicitly or implicitly) certain dependencies on neighbors, which may lead to loss of subtle but important structural information within the graph and other dependencies among neighbors. This intrigues us to ask the question: can we design a model to give the maximal flexibility of dependencies to each node's neighborhood. In this paper, we propose a novel graph node embedding (named PINE) via a novel notion of partial permutation invariant set function, to capture any possible dependence. Our method 1) can learn an arbitrary form of the representation function from the neighborhood, withour losing any potential dependence structures, and 2) is applicable to both homogeneous and heterogeneous graph embedding, the latter of which is challenged by the diversity of node types. Furthermore, we provide theoretical guarantee for the representation capability of our method for general homogeneous and heterogeneous graphs. Empirical evaluation results on benchmark data sets show that our proposed PINE method outperforms the state-of-the-art approaches on producing node vectors for various learning tasks of both homogeneous and heterogeneous graphs.

We propose a deep neural network framework for computing prices and deltas of American options in high dimensions. The architecture of the framework is a sequence of neural networks, where each network learns the difference of the price functions between adjacent timesteps. We introduce the least squares residual of the associated backward stochastic differential equation as the loss function. Our proposed framework yields prices and deltas on the entire spacetime, not only at a given point. The computational cost of the proposed approach is quadratic in dimension, which addresses the curse of dimensionality issue that state-of-the-art approaches suffer. Our numerical simulations demonstrate these contributions, and show that the proposed neural network framework outperforms state-of-the-art approaches in high dimensions.

A BSTRACT We present GQSA T, a branching heuristic in a Boolean SA T solver trained with value-based reinforcement learning (RL) using Graph Neural Networks for function approximation. Solvers using GQSA T are complete SA T solvers that either provide a satisfying assignment or a proof of unsatisfiability, which is required for many SA T applications. The branching heuristic commonly used in SA T solvers today suffers from bad decisions during their warm-up period, whereas GQSA T has been trained to examine the structure of the particular problem instance to make better decisions at the beginning of the search. Training GQSA T is data efficient and does not require elaborate dataset preparation or feature engineering to train. We train GQSA T on small SA T problems using RL interfacing with an existing SA T solver. We show that GQSA T is able to reduce the number of iterations required to solve SA T problems by 2-3X, and it generalizes to unsatisfiable SA T instances, as well as to problems with 5X more variables than it was trained on. We also show that, to a lesser extent, it generalizes to SA T problems from different domains by evaluating it on graph coloring. Our experiments show that augmenting SA T solvers with agents trained with RL and graph neural networks can improve performance on the SA T search problem. 1 I NTRODUCTION Boolean satisfiability (SA T) is an important problem for both industry and academia impacting various fields, including circuit design, computer security, artificial intelligence, automatic theorem proving, and combinatorial optimization. As a result, modern SA T solvers are well-crafted, sophisticated, reliable pieces of software that can scale to problems with hundreds of thousands of variables (Ohrimenko et al., 2009). SA T is known to be NPcomplete (Karp, 1972), and most state-of-the-art open-source and commercial solvers rely on multiple heuristics to speed up the exhaustive search, which is otherwise intractable. These heuristics are usually meticulously crafted using expert domain knowledge and are often iterated on using trial and error. In this paper, we investigate how we can use machine learning to improve upon an existing branching heuristic without leveraging domain expertise.

In order to learn effectively, robots must be able to extract the intangible context by which task progress and mistakes are defined. In the domain of reinforcement learning, much of this information is provided by the reward function. Hence, reward shaping is a necessary part of how we can achieve state-of-the-art results on complex, multi-step tasks. However, comparatively little work has examined how reward shaping should be done so that it captures task context, particularly in scenarios where the task is long-horizon and failure is highly consequential. Our Schedule for Positive Task (SPOT) reward trains our Efficient Visual Task (EVT) model to solve problems that require an understanding of both task context and workspace constraints of multi-step block arrangement tasks. In simulation EVT can completely clear adversarial arrangements of objects by pushing and grasping in 99% of cases vs an 82% baseline in prior work. For random arrangements EVT clears 100% of test cases at 86% action efficiency vs 61% efficiency in prior work. EVT + SPOT is also able to demonstrate context understanding and complete stacks in 74% of trials compared to a baseline of 5% with EVT alone. To our knowledge, this is the first instance of a Reinforcement Learning based algorithm successfully completing such a challenge. Code is available at https://github.com/jhu-lcsr/good_robot .

Locally adaptive activation functions with slope recovery term for deep and physics-informed neural networks Ameya D. Jagtap 1, Kenji Kawaguchi 2 and George Em Karniadakis 1,3, 1 Division of Applied Mathematics, Brown University, 182 George Street, Providence, RI 02912, USA. 2 Massachusetts Institute of T echnology, 77 Massachusetts Ave, Cambridge, MA 02139, USA. 3 Pacific Northwest National Laboratory, Richland, WA 99354, USA.Abstract We propose two approaches of locally adaptive activation functions namely, layer-wise and neuron-wise locally adaptive activation functions, which improve the performance of deep and physics-informed neural networks. The local adaptation of activation function is achieved by introducing scalable hyper-parameters in each layer (layer-wise) and for every neuron separately (neuron-wise), and then optimizing it using the stochastic gradient descent algorithm. Introduction of neuron-wise activation function acts like a vector activation function as opposed to the traditional scalar activation function given by fixed, global and layer-wise activations. In order to further increase the training speed, an activation slope based slope recovery term is added in the loss function, which further accelerate convergence, thereby reducing the training cost. For numerical experiments, a nonlinear discontinuous function is approximated using a deep neural network with layer-wise and neuron-wise locally adaptive activation functions with and without the slope recovery term and compared with its global counterpart. Moreover, solution of the nonlinear Burgers equation, which exhibits steep gradients, is also obtained using the proposed methods. On the theoretical side, we prove that in the proposed method the gradient descent algorithms are not attracted to sub-optimal critical points or local minima under practical conditions on the initialization and learning rate. Furthermore, the proposed adaptive activation functions with the slope recovery are shown to accelerate the training process in standard deep learning benchmarks using CIFAR-10, CIFAR-100, SVHN, MNIST, KMNIST, Fashion-MNIST, and Semeion data sets with and without data augmentation. Keywords: Machine learning, bad minima, stochastic gradients, accelerated training, PINN, deep learning benchmarks. 1. Introduction In recent years, research on neural networks (NNs) has intensified around the world due to their successful applications in many diverse fields such as speech recognition [13], computer vision [16], natural language translation [25], etc. Training of NN is performed on data sets before using it in the actual applications.

O PTIMAL T RANSPORT, C YCLEGAN, AND P ENALIZED LS FOR U NSUPERVISEDL EARNING IN I NVERSE P ROB-LEMS Byeongsu Sim 1 Gyutaek Oh 2 Sungjun Lim 2 Jong Chul Y e 1,2 1 Department of Mathematical Sciences, KAIST, Daejeon, Republic of Korea 2 Department of Bio and Brain Engineering, KAIST, Daejeon, Republic of Korea A BSTRACT The penalized least squares (PLS) is a classic approach to inverse problems, where a regularization term is added to stabilize the solution. Optimal transport (OT) is another mathematical framework for computer vision tasks by providing means to transport one measure to another at minimal cost. Cycle-consistent generative adversarial network (cycleGAN) is a recent extension of GAN to learn target distributions with less mode collapsing behavior. Although similar in that no supervised training is required, the algorithms look different, so the mathematical relationship between these approaches is not clear. In this article, we provide an important advance to unveil the missing link. Specifically, we reveal that a cycle-GAN architecture can be derived as a dual formulation of the optimal transport problem, if the PLS with a deep learning penalty is used as a transport cost between the two probability measures from measurements and unknown images. This suggests that cycleGAN can be considered as stochastic generalization of classical PLS approaches. Our derivation is so general that various types of cy-cleGAN architecture can be easily derived by merely changing the transport cost.

While neural networks have acted as a strong unifying force in the design of modern AI systems, the neural network architectures themselves remain highly heterogeneous due to the variety of tasks to be solved. In this chapter, we explore how to adapt the Layer-wise Relevance Propagation (LRP) technique used for explaining the predictions of feed-forward networks to the LSTM architecture used for sequential data modeling and forecasting. The special accumulators and gated interactions present in the LSTM require both a new propagation scheme and an extension of the underlying theoretical framework to deliver faithful explanations.