“Towards Efficient and Artificial General Intelligence using In-Memory Computing”
Tuesday, Jan. 23 at 1:00pm
LAR 234
Add to Calendar

Abstract

The tremendous progress within deep learning has hyper-fueled innovations within science and technology. This includes the discovery of algorithms to predict protein structures and enable interactive chatbots. With the end goal of making progress towards artificial general intelligence (the ability to independently reason and learn), deep learning systems are faced with computing-efficiency and generalization challenges. In the first part of this talk, Dr. Ewetz will provide an overview of his recent research on in-memory computing that aims to create future computing systems with significant (with orders of magnitude) higher energy efficiency. The key idea is to design a non-von Neumann computing system that avoids the costly shuttling of data between the CPU and memory. However, in-memory computing systems require cross-layer innovations across the computing stack to attain the potential advantages in performance. This includes the discovery of new computing paradigms, electronic design automation (EDA) algorithms, system designs, and algorithmic innovations. In the second part of the talk, he will provide an overview of his work on AI/ML that aims to enable AI systems to reason and learn independently. Dr. Ewetz will first cover his recent work on explainable AI that focuses on understanding the decision making of deep learning models. Next, he will provide an overview of his contributions to neuro-symbolic AI, which aims to combine the advantages of data driven learning (neural) with logical reasoning (symbolic). Finally, he will conclude the presentation with his future research vision and plans.

Biography

Dr. Rickard Ewetz is an associate professor in the Electrical and Computer Engineering (ECE) Department at the University of Central Florida. He received his Ph.D. degree in ECE from Purdue University in 2016. His research interests are broadly focused on the intersection of hardware and artificial intelligence. This includes creating electronic design automation algorithms, hardware/software co-design methodologies for emerging in-memory computing systems. He is actively working on AI/ML topics such as explainable AI, robust AI, and neuro-symbolic AI. Over the last ten years, he has published over 70 peer-reviewed articles, including 18 publications on the prestigious CS Ranking list (DAC, ICCAD, MICRO, AAAI, IJCAI). His research has received four best paper nominations from top-tier venues such as ICCAD, DATE, ASP-DAC, and MILCOM. His research is supported by DARPA, DOE, NSF, AFRL, Lockheed Martin Corp, Cyber-Florida, and the Florida High Tech Corridor Council. He has secured over $11M in external funding including a recent $2.7M project from DARPA on neuro-symbolic AI.