|Date and Time:||January 10 (Fri.), 2020, 13:00 – 14:30|
|Place:||Multimedia Hall #306, Building E-3(Map No.27), UEC|
|Speaker:||Vasileios Tserolas (Researcher, Center of Neuroscience and Biomedical Engineering, The University of Electro-Communications)|
|Chair:||Specially Appointed Prof. Shigeru TANAKA|
|Title:||Spiking neural networks as a paradigm of artificial intelligence|
|Abstract:||There is a long-lasting dream in creating artificial intelligence (AI). Today’s artificial intelligence is not yet there. The approach of today is to implement algorithms based on insights of human engineering. Hence, much effort is being invested into engineering new learning algorithms and information processing systems. The hope is that a right set of algorithms will be eventually created – making up a machine that will be able to learn on its own to the extent of becoming an AI. New general algorithms, once that could bring us closer to AI, do not seem to come out easily from such efforts. Some of the best general algorithms used today (e.g., deep learning) stem largely from 1980’s. The current AI technologies can do very well some things that require effort for a human (e.g., calculating prime numbers, searching databases) but have difficulties doing things that are for humans easy (e.g., perceiving, walking, navigating through space). So, what can we do? Is there any alternative or are we simply stuck with specialized AI?
For further developments of AI, we are in a need of using principles of biology to a higher degree than what we have been able to do so far. The new developments of AI must be based on a new theoretical approach on how biology and brain work resulting in a radically new view on what the nature of mental and cognitive operation is. Brains have evolved to control bodies in a very sophisticated way. Hence their abilities to quickly perceive the environment, rapidly detect statistical anomalies, control multiple degrees of freedom in real time, accumulate knowledge from many modalities, self-wire (learn) to optimize future behavior. When an animal navigates in the world, neurons can flexibly represent the position of the animal in a given environment, the composition of the environment, the head direction, the running speed, etc. These mental representations of the world are flexible and dynamic, determined and modulated by a range of environmental, behavioral, and neural parameters. In a similar way, to create biological-like AI it is necessary to mimic biology in respect to the levels of organization at which a biological agent adapts to its environment.
A good model problem for that is the control of an autonomous robot. Spiking neural networks (SNN) provide enormously rich expressing power but are not easy to control and very likely a pile of research will need to go into that field before first spiking chips will show amazing capabilities. Nevertheless, I believe that this is the right direction to go, in the long run.