Yasuo Tan

Computer Network Chair,
Japan Advanced Institute of Science and Technology,
Tel. +81-761-51-1159(secretary)
Tel. +81-761-51-1246(direct)
Fax. +81-761-51-1149

NOTE: Japanese page is here.

ytan's portrait

Research Interests

My research works are aimed at the development of next generation information environment, where countless hidden computers assist human life.

In that environment, computers must be more resilient; they must be more dependable under accidents including faults or misoperation, and it must be easier to construct a system for a specific application.

The Human Computer Interface, the style of a relationship between computers and human beings, is also one of the main issue for the information environment.

Real world oriented computing environment:

"Double clicking is hard enough to let my mom dislike Macs"

Each of computer systems currently used in our life has its own virtual world in it. Although these worlds have user interfaces which attempt to mimic the real world, they are distinct and disjoint from the real one, and users must interact with both of these worlds.

For better computer-human relationship, there is an approach called ubiquitous computing or embodied virtuality where computers are taken into or embedded into the real world.

We are trying to make computer systems which support intelligent work of human beings in a way that machines act as hyper-real objects, not virtual objects based on metaphor of real things.

As the activity on this project, we are currently developing a plug and play video LAN system and location information handling architecture.

Fault-tolerance in neural networks:

"Neurons in the human brain die every day without affecting its performance. Why can't machines?"

Although it is widely believed that neural networks should have a capability of fault-tolerance because a brain of animals can be considered fault-tolerant, models currently used in engineering fields do not always show this property. We have developed some schemes to bring out a potential fault-tolerant ability of multi-layer neural networks. Learning based approach, where a measure of fault-tolerance is introduced into the objective function of learning, enabled machines that acquire fault-tolerance with no concrete technique for fault-tolerance given. It is also shown that there is a close relationship between fault-tolerance and generalization ability.

One of the most promising technologies for system construction is the Genetic Algorithm (GA), an engineering model of the evolution of life. We have proposed a GA based learning algorithm for multi-layer neural networks, through which better networks are produced in a viewpoint of fault-tolerance as well as performance against given tasks.

Professor Yasuo Tan was born in Hokkaido Pref. in 1965. He came to JAIST in April, 1993 from Tokyo Institute of Technology. He recieved his Ph.D from Tokyo Institute of Technology.

He is a member of IEEJ, IPSJ, IEICE, JNNS and JSSST.

Selected Publications

"Scaling up IEEE1394 DV network to an enterprise video LAN with ATM technology"; Proc. IEEE ICCE 98, 1998

"LIBRA - Location Information BRoker Architecture"; IPSJ Tech. report, 97-MBL-2-2, 1997

"Document distribution system in the ubiquitous computing environment"; Proc. IPSJ Programming Symposium, 1995

"A fault-tolerant multilayer neural network model and its properties"; Systems and Computers in Japan, Vol. 25, No. 2, 1994.

"Fault-tolerant back-propagation model and its generalization ability"; Proc. IJCNN93, Vol. 3, pp. 2516-2519, 1993

"A network with multi-partitioning units"; Proc. IJCNN89, Vol. 2, pp. 439-442, 1989

Publication List

Invited Talks

Current Research

Plug-and-play Video-LAN architecture and protocols.

Location Information service architecture.

Multi-threaded processor architecture