We explore an approach to behavioral cloning in video games. We are motivated to pursue a learning architecture that is data efficient and provides opportunity for interpreting player strategies and replicating player actions in unseen situations. To this end, we have developed a generative model that learns latent features of a game that can be used for training an action predictor. Specifically, our architecture combines a Variational Autoencoder with a discriminator mapping the latent space to action predictions (predictor). We compare our model performance to two different behavior cloning architectures: a discriminative model (a Convolutional Neural Network) mapping game states directly to actions, and a Variational Autoencoder with a predictor trained separately. Finally, we demonstrate how we can use the advantage of generative modeling to sample new states from the latent space of the Variational Autoencoder to analyze player actions and provide meaning to certain latent features.

The variational quantum Monte Carlo (VQMC) method received significant attention in the recent past because of its ability to overcome the curse of dimensionality inherent in many-body quantum systems. Close parallels exist between VQMC and the emerging hybrid quantum-classical computational paradigm of variational quantum algorithms. VQMC overcomes the curse of dimensionality by performing alternating steps of Monte Carlo sampling from a parametrized quantum state followed by gradient-based optimization. While VQMC has been applied to solve high-dimensional problems, it is known to be difficult to parallelize, primarily owing to the Markov Chain Monte Carlo (MCMC) sampling step. In this work, we explore the scalability of VQMC when autoregressive models, with exact sampling, are used in place of MCMC. This approach can exploit distributed-memory, shared-memory and/or GPU parallelism in the sampling task without any bottlenecks. In particular, we demonstrate GPU-scalability of VQMC for solving up to ten-thousand dimensional combinatorial optimization problems.

We present a fast, petaflop-scalable algorithm for Stokesian particulate flows. Our goal is the direct simulation of blood, which we model as a mixture of a Stokesian fluid (plasma) and red blood cells (RBCs). Directly simulating blood is a challenging multiscale, multiphysics problem. We report simulations with up to 200 million deformable RBCs. The largest simulation amounts to 90 billion unknowns in space. In terms of the number of cells, we improve the state-of-the art by several orders of magnitude: the previous largest simulation, at the same physical fidelity as ours, resolved the flow of O(1,000-10,000) RBCs. Our approach has three distinct characteristics: (1) we faithfully represent the physics of RBCs by using nonlinear solid mechanics to capture the deformations of each cell; (2) we accurately resolve the long-range, N-body, hydrodynamic interactions between RBCs (which are caused by the surrounding plasma); and (3) we allow for the highly non-uniform distribution of RBCs in space. The new method has been implemented in the software library MOBO (for “Moving Boundaries”). We designed MOBO to support parallelism at all levels, including inter-node distributed memory parallelism, intra-node shared memory parallelism, data parallelism (vectorization), and fine-grained multithreading for GPUs. We have implemented and optimized the majority of the computation kernels on both Intel/AMD x86 and NVidia's Tesla/Fermi platforms for single and double floating point precision. Overall, the code has scaled on 256 CPU-GPUs on the Teragrid's Lincoln cluster and on 200,000 AMD cores of the Oak Ridge national Laboratory's Jaguar PF system. In our largest simulation, we have achieved 0.7 Petaflops/s of sustained performance on Jaguar.

We present fast adaptive parallel algorithms to compute the sum of N Gaussians at N points. Direct sequential computation of this sum would take O(N2) time. The parallel time complexity estimates for our algorithms are O (N/np) for uniform point distributions and O (N/nplog N/+ nplog np) for nonuniform distributions using npCPUs. We incorporate a planewave representation of the Gaussian kernel which permits "diagonal translation". We use parallel octrees and a new scheme for translating the plane-waves to efficiently handle nonuniform distributions. Computing the transform to six-digit accuracy at 120 billion points took approximately 140 seconds using 4096 cores on the Jaguar supercomputer at the Oak Ridge National Laboratory. Our implementation is kernel-independent and can handle other "Gaussian-type" kernels even when an explicit analytic expression for the kernel is not known. These algorithms form a new class of core computational machinery for solving parabolic PDEs on massively parallel architectures.