We propose InSituNet, a deep learning based surrogate model to support parameter space exploration for ensemble simulations that are visualized in situ. In situ visualization, generating visualizations at simulation time, is becoming prevalent in handling large-scale simulations because of the I/O and storage constraints. However, in situ visualization approaches limit the flexibility of post-hoc exploration because the raw simulation data are no longer available. Although multiple image-based approaches have been proposed to mitigate this limitation, those approaches lack the ability to explore the simulation parameters. Our approach allows flexible exploration of parameter space for large-scale ensemble simulations by taking advantage of the recent advances in deep learning. Specifically, we design InSituNet as a convolutional regression model to learn the mapping from the simulation and visualization parameters to the visualization results. With the trained model, users can generate new images for different simulation parameters under various visualization settings, which enables in-depth analysis of the underlying ensemble simulations. We demonstrate the effectiveness of InSituNet in combustion, cosmology, and ocean simulations through quantitative and qualitative evaluations.

We investigate the representation of discrete scientific data with a Ckfunctional model, where Ckdenotes k-th order continuity, in a distributed-memory parallel setting. The Multivariate Functional Approximation (MFA) model is a piecewise-continuous functional approximation based on multi-variate high-dimensional B-splines. When computing an MFA approximation in parallel over multiple blocks in a spatial domain decomposition, the interior of each block will be Ck, k being the B-spline polynomial degree, but discontinuities exist across neighboring block boundaries. We present an efficient and scalable solution that involves blending neighboring approximations to ensure Ckcontinuity across block boundaries. We show that after decomposing the domain in structured, overlapping blocks and approximating blocks independently to high degrees of accuracy, we can extend the local solution, in a postprocessing step, to the global domain by using compact, multidimensional smoothstep functions. We prove that this approach, which can be viewed as an extended partition of unity approximation method, is scalable on high-performance computing architectures.

Ptychography is an imaging technique which aims to recover the complex-valued exit wavefront of an object from a set of its diffraction pattern magnitudes. Ptychography is one of the most popular techniques for sub-30 nanometer imaging as it does not suffer from the limitations of typical lens based imaging techniques. The object can be reconstructed from the captured diffraction patterns using iterative phase retrieval algorithms. Over time many algorithms have been proposed for iterative reconstruction of the object based on manually derived update rules. In this paper, we adapt automatic differentiation framework to solve practical and complex ptychographic phase retrieval problems and demonstrate its advantages in terms of speed, accuracy, adaptability and generalizability across different scanning techniques.

It is well established that the human brain outperforms current computers, concerning pattern recognition tasks, through the collaborative processing of simple building units (neurons). In this work we expand an abstracted model of the neocortex called Hierarchical Quilted Self Organizing Map, benefiting from the parallel power of current Graphical Processing Units, to achieve realtime understanding and classification of spatio-temporal sensory information. We also propose an improvement on the original model that allows the learning rate to be automatically adapted according to the input training data available. The overall system is tested on the task of gesture recognition from a Microsoft Kinect publicly available dataset.

This paper describes our latest implementation of Particle Swarm Optimization (PSO) with simple ring topology for modern Graphic Processing Units (GPUs). To achieve both the fastest execution time and the best performance, we designed a parallel version of the algorithm, as fine-grained as possible, without introducing explicit synchronization mechanisms among the particles' evolution processes. The results we obtained show a significant speed-up with respect to both the sequential version of the algorithm run on an up-to-date CPU and our previously developed parallel implementation within the nVIDIA CUDA architecture.