Title: Two Forms of Immediate Reward Reinforcement Learning for Exploratory Data AnalysisSpeaker:Pei Ling Lai, Professor of Southern Taiwan University (Dept. of Electronics Engineering)pllai@mail.stut.edu.twDate: 2:00PM ~ 3:00PM, Jan 31(Thursday), 2008Place: ICP lecture room, Science Library 6F (과도관 6층 ICP 강의실)We review two forms of immediate reward reinforcement learning: in thefirstthe learner is a stochastic node and the second the individualunit is deterministic but has stochastic synapses. We illustrate thefirst method on the problem of Independent Component Analysis. Fourlearning rules are developed from the second perspective and their usefor learning rules performing linear projection techniques such asprincipal component analysis, exploratory projection pursuit andcanonical correlation analysis is investigated. The method is verygeneral simply requiring a reward specific to the function we requirethe unit to perform. We also discuss how to use the method forlearning kernel mappings and conclude by illustrating its use on atopology preserving mapping.