Prof. Naoshi Sakamoto
Tokyo Denki University, Japan

Naoshi Sakamoto holds M.S. and Dr.S. degree from Tokyo Institute of Technology. Currently, he is a professor in Tokyo Denki University, Japan. Since 1992, he had been an assistant researcher in Hitotsubashi University and Tokyo Institute of Technology, successively. He had engaged in constructing and managing the campus area network. Originally, he was studying theoretical computer science. He won IEICE excellent paper award in 1999 for the study of the amount of information among initial conditions for distributed algorithms. Since 2001, he has been an associate professor at Tokyo Denki University. Then, since 2014, he has been a professor. Since 2016, he has also been an deputy director of Office of Educational Development for two years in the University.  

Speech Title: Learning Complexity of Neural Networks
Abstract: Recently, deep learning is developing greatly. Particularly, the field of image recognition and the field of machine translation have been making remarkable progress.
Moreover, there are various applications in many fields such as the field of information technology. Now, many researchers can not ignore them.
A current deep neural network requires a lot of input data, high spec computer, and long training time. Then, it acquires better recognizing ability than the human recognizing ability.
Now, when a neural network learns, how much does it require computational power and training time? The learning complexity of neural networks has been studying for a long time. The speaker is introducing the fundamental theories and recent results about learning of neural networks.