Sudhir Kylasa graduated with a PhD from Purdue University in December 2019. His PhD thesis is based on the application of higher order solvers in machine learning applications and this codebase is publicly available to download which includes a CUDA framework for computing Hessian-vector products and estimating fisher-blocks for fully-connected layers and convolution layers. His publications are featured in several reputed journals and conferences. His interests span across machine learning, deep learning, distributed computing, operating systems and embedded systems with a particular interest towards GPU acceleration of the underlying optimization problems. Prior to pursuinghis doctoral degree Dr. Kylasa worked in software industry at various capacities.
Accelerators, FPGA, and GPUs