Keiji Kimura

Waseda University, Japan

Evaluation of Deep Predictive Learning for AI-driven Robots on an Edge-GPU

Abstract

Expectations for AI-driven robots are increasing, particularly in the medical and nursing fields. Deep predictive learning technology, which predicts the robot’s future behavior using camera images and sensor data, has been proposed to realize such intelligent and flexible robots. The robots’ bodies must be compact to realize their working in people’s living fields. Thus, energy-efficient AI processors that do not require large battery and cooling modules are indispensable. While quantization has been a popular technique to improve performance and energy efficiency for image recognition, its significance for deep predictive learning is unclear since it uses information other than image inputs. This talk will show the evaluation result of deep predictive learning with quantization on an Nvidia Jetson Orin Nano in terms of performance, accuracy, and energy consumption. Then, we will discuss desirable AI processors for deep predictive learning.

Biography

Keiji Kimura received a Ph.D. in electrical engineering from Waseda University in 2001. He was an assistant professor in 2004, an associate professor in 2005, and a professor in 2012 at Waseda University. He has been a director of the Green Computing System Research Organization in Waseda since 2019. He received the 2014 MEXT (Ministry of Education, Culture, Sports, Science and Technology in Japan) award. His research interests include multicore processor architecture and parallelizing compiler technologies. He is a member of IPSJ, ACM, and IEEE. He has served on the program committee of many conferences.