Situational Descriptions of Behavioral Procedures: The in Situ Testbed

Abstract

We demonstrate the In Situ testbed, a system that aids in evaluating computational models of learning, including artificial neural networks. The testbed models contingencies of reinforcement rising an extension of Mechner's (1959) notational system for the description of behavioral procedures. These contingencies are input to the model under test. The model's output is displayed as cumulative records. The cumulative record can then be compared to one produced by a pigeon exposed to the same contingencies. The testbed is tried with three published models of learning. Each model is exposed to up to three reinforcement schedules (testing ends when the model does not produce acceptable cumulative records): continuous reinforcement and extinction, fixed ratio, and fixed interval. The In Sitt testbed appears to be a reliable and valid testing procedure for comparing models of learning.

Simulating a shaping task.

“…The main result of both the experiment and the simulation is that keypecking successfully tracked We refer to the present simulation as "semisituational," a term that deserves some clarification. A "fully situated" model is an ideal (Kemp & Eckerman, 2001;Kemp, in press). In practice, we say that a model is "situated" within the testbed if the environment and organism are modeled in sufficient detail such that organismic inputs and outputs can be unambiguously identified with their environmental causes and effects.…”

“…Testbed design. The overall structure of the testbed conforms to the In Situ design (Kemp & Eckerman, 2001), consisting of a virtual reality (VR) component to model the environmental contingencies, an artificial life (AL) component to model the sensorimotor capacities of the organism, and an artificial intelligence (AI) component to model the behavioral functions. The AI component corresponds to the neural network model being tested.…”

The Emergence of Stimulus Relations: Human and Computer Learning

“…For example, Donahoe and Burgos (2000) employed neural network training to identify levels of response strength in terms of accelerated and extinguished operant behavior. Kemp and Eckerman (2001) continued the lower organism learning analogy by exposing well-trained algorithms to various schedules of reinforcement and obtained cumulative records with response patterns very similar to those of nonverbal organisms exposed to the same schedules of reinforcement. Burgos (2007) demonstrated that neural networks were capable of performing in ways that were analogous to autoshaping and automaintenance (cf.…”

Neural Network and Multivariate Analyses: Pattern Recognition in Academic and Social Research

“…For example, Donahoe and Burgos (2000) described a series of feedforward neural network manipulations utilized to assess simulated response strength to feedback in terms of decreasing and increasing operant behavior. Kemp and Eckerman (2001) advanced a variation on this theme by exposing a feedforward network to various schedules of reinforcement, and they obtained cumulative records showing response patterns reminiscent of nonverbal organisms on the same schedules. More recently, Burgos (2007) developed a series of four simulations using feedforward architectures to demonstrate autoshaping and automaintenance.…”