Professor Sungjoo Yoo and Professor Byung-Gon Chun of Seoul National Univesrity Department of Computer Science and Engineering won the Facebook Caffe2 Research Award


▲Professor Sungjoo Yoo (left), Professor Byung-Gon Chun(right) of Seoul National Univesrity Department of Computer Science and Engineering
 

Seoul National University College of Engineering (Dean Kookheon Char) announced on 11^th that Professor Sungjoo Yoo and Professor Byung-Gon Chun of Department of Computer Science and Engineering were selected as candidates for the Facebook Caffe2 Research Award program in September this year.
Caffe2 is an open source deep-learning project under development on Facebook, and it provides a variety of platforms from cluster environment to mobile environment by utilizing many GPUs. Facebook provides an opportunity to work closely with Facebook while selecting and supporting research institutions that can perform excellence research in the field of deep learning through the program.
 
Professor Sungjoo Yoo is working on a neural network energy and performance optimization study on mobile devices and 4-bit data quantization research. Since the low-power deep-learning implementations are essential for mobile devices such as smart phones and glass, quantization techniques that reduce the size of data can improve both energy efficiency per unit of motion and computational power per unit area. Currently, the highest level of quantization is possible to use 8-bit data for large-scale neural networks, and Professor Yoo's goal is to develop a optimization technology that reduces the data size to 4-bit level.
 
A research team led by Professor Byung-Gon Chun is conducting large-scale deep-learning and ultra-fast deep-learning inference research. Models developed with a deep-learning framework such as Caffe2 usually run on a single server and a single GPU. These models can be quickly learned on multiple GPUs and multiple servers using a large volume of data. The team aims to develop a technology that automatically distributes Caffe2's single device and single GPU execution model in consideration of cluster environment, data, and model complexity.
 
Professor Yoo and Professor Chun said, "We are very pleased to be selected as a Facebook Caffe2 Research Award. We will continuously endeavor to achieve innovative research results. "