초청강연 Invited Speech

Beyond Perception and Toward Cognition

이민호 교수

경북대학교 전자공학부

Biography

Minho Lee received the Ph.D. from KAIST in 1995, and is currently a professor of School of Electronics Engineering and a director for KNU-LG Electronics Convergence Research Center and Artificial Intelligence Research Center, Kyungpook National University, Taegu, Korea. He established Mobile Technology Commercial Center at Daegu, and worked for Education & Training Department as a director from 2005 to 2006. Also, he was a visiting scholar for Dept. of Brain and Cognitive Science at MIT from 2006 to 2007. He was president for Asia-Pacific Neural Network Assembly (APNNA) at 2013. He received several awards such as APNNA Excellent Service Award (2014) and Best Industry-Academic Cooperation Award (2014), and best paper awards at international conferences including AEARU (2015), ICONIP (2007 and 2009), IDEAL(2008), ICAISC(2006) and so on. He has been served for several international journals (Neural Networks, Neural Processing Letters, and Nature Intelligence, etc.) as an associate editor and also for international conference as general chairs for ICONIP2013 and HAI2015 and program chairs for ICONIP2009 and ICONIP2016. His research interests include deep neural networks, brain-neuroinformatics, biologically inspired vision systems, human augmented cognition, selective attention, brain-machine interaction and intelligent sensor systems. (Homepage: http://abr.knu.ac.kr)

Abstract

In recent years, deep learning, which is a machine learning algorithm for neural networks with many hidden layers, has shown the state-of-the-art performance in many engineering applications including image recognition, video understanding, speech recognition and natural language processing such as translation and summarization. The deep learning methods have taken a place not only in many professional academic areas but also in industrial arena. World’s leading companies like Google, Microsoft and Facebook have teams of specialists working on deep learning, and publish their research in reputed journals. The primary question is: what is the main difference between conventional approaches and deep learning methods? Actually, there already had been several approaches like deep learning but they were impractical. However, recent breakthroughs have been introducing and adapting, deep learning methods eventually defeating many classical state-of-art algorithms in many engineering application fields. Moreover, they have shed light to the brain-like artificial intelligence. In this talk, I will briefly explain main characteristics and issues of conventional machine learning and deep learning methods. Then, I will introduce research works at my laboratory in this area, which includes to design the optimal number of deep hidden layers, handling incomplete and long sequence data, speeding up the training process. Furthermore, I will explain the current research in my lab, focusing on the development of deep cognitive neural networks for emotion and intention understanding, including using perception-action cycle learning to implement a brain augmented autonomous cognition system. Real world applications based on the developed models will be briefly introduced.