Synthesis Lectures on Engineering, Science, and Technology

This book describes an extensive and consistent soft error assessment of convolutional neural network (CNN) models from different domains through more than 14.8 million fault injections, considering different precision bit-width configurations, optimization parameters, and processor models. The authors also evaluate the relative performance, memory utilization, and soft error reliability trade-offs analysis of different CNN models considering a compiler-based technique w.r.t. traditional redundancy approaches.

The monograph is devoted to the construction of the high-order finite difference and finite element methods for numerical solving multidimensional boundary-value problems (BVPs) for different partial differential equations, in particular, linear Helmholtz and wave equations, nonlinear Burgers’ equations, and elliptic (Schrödinger) equation. Despite of a long history especially in development of the theoretical background of these methods there are open questions in their constructive implementation in numerical solving the multidimensional BVPs having additional requirement on physical parameters or desirable properties of its approximate solutions.

Over the last two decades many papers on this topics have been published, in which new constructive approaches to numerically solving the multidimensional BVPs were proposed, and its highly desirable to systematically collect these results. This motivate us to write thus monograph based on our research results obtainedin collaboration with the co-authors. Since the topic is importance we believe that this book will be useful to readers, graduate students and researchers interested in the field of computational physics, applied mathematics, numerical analysis and applied sciences