1. 제 목 : Intrinsic, Autonomous Semantics in Natural and Artificial
Agents.
2. 발표자 : Yoonsuck Choe (Taxas A&M University 컴퓨터과학과 조교수)
3. 일 시 : 2005년 11월 24일 (목) 10시～12시
4. 장 소 : 경북대 공대10호관 613호
5. 초청교수 : 이 민 호 교수
6. 강사약력
- 1993. 연세대학교 컴퓨터과학과 학사 졸업
- 1995 ~ 2001. University of texas at Austin 석사․박사 졸업
- 현재 ~ Texas A & M University 컴퓨터과학과 조교수
7. 내용요약 :
How can we build artificial agents that can autonomously explore andunderstand their environment? An immediate requirement for such an agentis to understand what its own internal sensory state means, i.e., tolearnthe semantics of its internal state. In principle, we as designers canprovide the agents with the required semantics, but this turns out tobe achallenging engineering problem. To overcome this challenge, we need tofirst look at natural agents and see how they acquire meaning of theirownsensory states--their neural firing patterns. We can learn a lot aboutwhat certain neural spikes mean by carefully controlling the inputstimulus and observing how the neurons fire. However, neurons embeddedinthe brain do not have direct access to the outside stimuli, so such astimulus-to-spike association may not be possible within the brain. Howthen does the brain solve this problem? We propose that motor action isnecessary to overcome this conundrum. Further, we provide a simple yetpowerful criterion: sensory-invariance, for learning the meaning ofsensory states. The basic idea is that a particular action sequence,whichmaintains invariance of a certain sensory state, reflects importantproperties of the environmental stimulus that triggered the sensorystate.Our experiments with a simplified sensorimotor agent shows thatsensory-invariance can indeed serve as a powerful objective for learningsemantics. By understanding principles underlying sensorimotorgrounding,we expect to be able to build agents that have a more autonomousunderstanding of its environment.
* This is a joint work with S. Kumar Bhamidipati and Stuart B. Heinrich.
8. 주최 : BK21 정보기술연구사업단, 전자전기컴퓨터학부
|
