Abstract: A local buffer is integrated with a witness generator and a proof generator on a cryptographic processor and is separate from host memory accessed by a host processor operating with the cryptographic processor in a proving computing system. The witness generator: receives, from software program running on the host processor, compiled code of a zero-knowledge-proof (ZKP) program and specific input to the ZKP program; executes the ZKP program by way of executing the compiled code; records specific output generated from the ZKP program with the specific input, intermediate variable values, and the specific input, as a specific witness of executing the ZKP program; stores the specific witness in the local buffer. The proof generator: receives, from the software program running on the host processor, a proving key; accesses the specific witness in the local buffer; generates a specific zero-knowledge proof for executing the ZKP program with the specific input.

Abstract: System and methods to train a neural network to systematically find a cross-over point, given the number of devices (e.g., Graphical Processing Units) used to train a deep learning (DL) model, that indicates which parallelization strategy to implement when optimizing the training of the DL model on a particular system to achieve maximum efficiency gains.