Mike Kemelmakher
About Mike Kemelmakher
Mike Kemelmakher is the Head of ML/DL Acceleration and Performance Optimization at Vianai Systems, Inc., and a co-founder of ArgosMD, Inc. He has extensive experience in machine learning, having held various leadership roles at SAP and worked on optimizing inference for advanced models.
Current Role at Vianai Systems
Mike Kemelmakher serves as the Head of ML/DL Acceleration and Performance Optimization at Vianai Systems, Inc. since 2021. In this role, he is responsible for leading efforts to optimize machine learning and deep learning models, focusing on performance and efficiency. His work involves engaging in hardware and software co-design to enhance the execution of model graphs on various hardware accelerators. He operates in a hybrid environment, splitting his time between Israel and California.
Previous Experience at SAP
Prior to his current position, Mike Kemelmakher held multiple roles at SAP. He worked as the Director of System Architecture for In-Memory Data and Analytic Engines (HANA) from 2011 to 2012. He later served as the Head of the SAP Innovation Center in Israel from 2014 to 2016 and as Head of AI, Computer Vision applied R&D from 2016 to 2018. Additionally, he was the Head of the HANA Co-Innovation Lab from 2012 to 2014, contributing to various innovative projects within the organization.
Educational Background
Mike Kemelmakher completed his B.Sc. in Computer Science and Statistics at Bar-Ilan University. He further pursued advanced studies, achieving an M.Sc. in Computer Science with a focus on Distributed Systems and High-Performance Computing (HPC). He participated in the MOSIX Project during his studies. Additionally, he completed two years of the Ph.D. program in Computer Science at Bar-Ilan University, specializing in distributed systems and autonomous robots.
Technical Expertise and Research Focus
Mike Kemelmakher leads a team that specializes in optimizing inference for various deep learning models, including convolutional neural networks (CNNs) and transformer-based models. His expertise includes low-level model execution analysis and profiling across different hardware platforms such as Nvidia, Intel, QAIC, and ARM. He employs techniques like tensor decomposition, pruning, quantization, operator fusion, and linear attention approximation for model acceleration. His recent work has concentrated on LLM use-cases, including serving optimization on runtimes like Nvidia TensorRT LLM and HF TGI.
Career in Academia and Early Experience
Before his extensive industry experience, Mike Kemelmakher worked as a Lecturer in the Computer Science Department at The Open University of Israel from 2005 to 2016. He also served as a Research Assistant at Bar-Ilan University from 1996 to 2001. His early career included roles at IBM Israel as a UNIX Expert and HPC Expert, as well as serving in the Israel Defense Forces as an Infrastructure Team Leader and System Engineer from 1998 to 2001.