Computational Models of Conditioning

Abstract

Since first described, multiple properties of classical conditioning have been discovered, establishing the need for mathematical models to help explain the defining features. The mathematical complexity of the models puts our understanding of their workings beyond the ability of our intuitive thinking and makes computer simulations irreplaceable. The complexity of the models frequently results in function redundancy, a natural property of biologically evolved systems that is much desired in technologically designed products. Experts provide the latest advancements in the field and present detailed descriptions of how the models simulate conditioned behaviour and its physiological bases. It offers advanced students and researchers examples of how the models are used to analyse existing experimental results and design future experiments. This volume is of great interest to psychologists and neuroscientists, as well as computer scientists and engineers searching for ideas applicable to the design of robots that mimic animal behaviour.

What Have Computational Models Ever Done for Us?

“…Hence, it is widely accepted that classical conditioning is at the basis of most learning phenomena and behavior and thus paramount that we develop accurate models of conditioning. In this endeavor, collaboration between psychologists and computer scientists has enjoyed considerable success (Schmajuk, 2010a;Schmajuk, 2010b;Alonso & Mondragón, 2011). This collaboration is sustained on well-known arguments: Expressing models as sets of algorithms grants us formal ways of representing psychological intuitions and means of calculating their predictions accurately and quickly; from computational models we also borrow a view, the so-called computer metaphor, on how information is processed that has proved useful in understanding cognition; moreover, the architectures in which computational models are implemented, artificial neural networks for instance, resemble those of associative learning, both at conceptual and neural levels; finally, machine learning models, such as temporal difference learning and Bayesian learning, can be viewed as effective abstractions of how associations are formed and processed.…”