Deep Learning (DL) algorithms are increasingly significant in solving problems of practical scale and utility. These breakthroughs continue at the intersection of both hardware and software: New model architectures as well hardware accelerators are being created by the month across several companies. As Moore’s law and Dennard’s scaling give diminishing returns, this simultaneous optimization of hardware and software is essential to continue to reduce the cost and energy of DL systems. We will give an overview of this space and discuss in detail one axis of such co-design. We will also share briefly the roadmap of building an accelerator for the open-source SHAKTI family of processors we are developing at IIT Madras.

He received Bachelors and Masters of Technology in Electrical Engineering from IIT Bombay in Aug 2009. He completed his PhD in Computer Engineering from ETH Zurich in Jan 2014, under the supervision of Prof. Lothar Thiele. He was a post-doc at ETH Zurich from Jan to Dec 2014. I then spent over 2.5 years at IBM Research, Bangalore and a few months consulting for machine learning startups. He joined IIT Madras as an Assistant Professor in Mar 2018.