Two-mode Threshold Learning
Report Number: AMRL TDR 64-39
Author(s): Sklansky, J.
Corporate Author(s): RCA Laboratories
Laboratory: Biophysics Laboratory
Date of Publication: 1964-05
Pages: 53
Contract: AF 33(657)-11336
DoD Project: 7233 - Biological Information Handling Systems and Their Functional Analogs
DoD Task: 723305 - Theory of Information Handling
Identifier: AD0602966
Abstract:
In certain 'threshold learning processes' (TLPs) associated with pattern recognition and sensory perception, the process of training an observer to recognize patterns or distinguish levels of sensory excitation may be modeled by a finite-state Markov chain. The statistics of the signals received by the observer move at random between two sets of parameters in a 'two-mode' TLP, modeled by a two-mode Markov chain. Using a probabilistic measure of effectiveness, the effectiveness of a 'simple incremental' feedback policy is shown to be greater for two-mode TLPs than for one-mode TLPs over a certain range of environmental and structural statistics. A method of designing periodic train-work schedules for two-mode TLPs is described. ('Train' and 'work' correspond to 'closed-loop' and 'open-loop' respectively.) In many real adaptive processes an 'RC approximation' of the train-work dynamics is applicable. For these processes the ratio of working time to retraining time, yielding a desired performance level, is maximized when the work-retrain period is made as small as possible. Many stochastic processes present modeling problems of near psychological complexity. Ways in which open-loop/closed-loop relationships can help the life scientist or engineer model adaptive stochastic processes by two-mode TLPs are indicated.
Provenance: RAF Centre of Aviation Medicine
Author(s): Sklansky, J.
Corporate Author(s): RCA Laboratories
Laboratory: Biophysics Laboratory
Date of Publication: 1964-05
Pages: 53
Contract: AF 33(657)-11336
DoD Project: 7233 - Biological Information Handling Systems and Their Functional Analogs
DoD Task: 723305 - Theory of Information Handling
Identifier: AD0602966
Abstract:
In certain 'threshold learning processes' (TLPs) associated with pattern recognition and sensory perception, the process of training an observer to recognize patterns or distinguish levels of sensory excitation may be modeled by a finite-state Markov chain. The statistics of the signals received by the observer move at random between two sets of parameters in a 'two-mode' TLP, modeled by a two-mode Markov chain. Using a probabilistic measure of effectiveness, the effectiveness of a 'simple incremental' feedback policy is shown to be greater for two-mode TLPs than for one-mode TLPs over a certain range of environmental and structural statistics. A method of designing periodic train-work schedules for two-mode TLPs is described. ('Train' and 'work' correspond to 'closed-loop' and 'open-loop' respectively.) In many real adaptive processes an 'RC approximation' of the train-work dynamics is applicable. For these processes the ratio of working time to retraining time, yielding a desired performance level, is maximized when the work-retrain period is made as small as possible. Many stochastic processes present modeling problems of near psychological complexity. Ways in which open-loop/closed-loop relationships can help the life scientist or engineer model adaptive stochastic processes by two-mode TLPs are indicated.
Provenance: RAF Centre of Aviation Medicine