Abstract
Scaling up deep learning is becoming increasingly crucial for large-scale applications, where the available data as well as the models are typically very large. Distributed deep learning refers to a class of algorithms that are focused on learning from data distributed among multiple computational sources such as cluster nodes, multiple networked computers, edge devices, and robots. Within the class of distributed deep learning algorithms, decentralized deep learning algorithms form a subset that aims to learn global models by leveraging peer-to-peer communication of metadata among the learning agents without transmitting their private data sets. In this talk, I will discuss a few aspects of decentralized deep learning that we explored over the past few years such as performance, scalability, robustness to non-IID data distribution and communication efficiency.
Biography
Dr. Soumik Sarkar received his Ph.D. in Mechanical Engineering from Penn State in 2011. Currently he is an Associate Professor of Mechanical Engineering and Computer Science, a Walter W Wilson Faculty Fellow in Engineering and the Director of the Translational AI Center at Iowa State. He also serves as the Associate Director of the USDA-NIFA sponsored AI Institute for Resilient Agriculture (AIIRA). Dr. Sarkar’s research interests include Machine Learning and Decision & Control with applications to Cyber-Physical Systems such as energy, transportation, design & manufacturing and agriculture systems. He co‐authored more than 240 peer-reviewed publications and received about $52M research funding over the past seven years. Dr. Sarkar is a recipient of the prestigious Young Investigator award from the US Air Force Office of Scientific Research (AFOSR) in 2017 and the NSF CAREER award in 2019.
