|InterJournal Complex Systems, 248
|Manuscript Number: |
Submission Date: 981216
|Effect of History to Achieve Self Adaptation|
Subject(s): CX.66, CX.13, CX.16, CX.17
Category: Review Article
Many neural networks have been designed sucessfully to make a temporal associations between stimulus and response. However, once this association is made there is no inherent mechanism for it to be changed and the system becomes determined. This has an advantage of efficiency of producing a predictable output to a given input at the expense of its effectiveness, that is, adaptability to choose appropriate responses as its environment changes over time. In our previous work on temporal associations we had used Spatio-TEmporal Neurons (STEN) and an Adaptive Threshold Learning (ATL). STEN is sensitive to the spatio-temporal changes in the input stream of a signal. ATL uses time-varying and learning threshold functions and a forgetting constant among other things. This provides it temporal sensitivity to simulate various aspects of temporal associations.In addition it provides a trace for the systemís history that is explored in the current work. In this work we are proposing ATL that can by keeping track of its history in a manner that allows the system to adapt to the changes in its environment. In other words, we are creating a situation, through self-reflection, in which even the previously conditioned system has a moment of choice, between its stimulus and response, where it can break its previously learned associations. This allows this neural system to be able to be self directed and change its behavior to a previously learned response. Thus, it has the freedom to choose to adapt to new situations. Hence we are able to create intelligent systems that can move from deterministic to a new behavior where the past influences the future but does not necessarily determine it.
|Submit referee report/comment|