A hybrid neural network model based reinforcement learning agent
ISNN'10 Proceedings of the 7th international conference on Advances in Neural Networks - Volume Part I
The evolution of representation in simple cognitive networks
Neural Computation
Introduction to "The Material Bases of Cognition"
Minds and Machines
Minds and Machines
| Hi-index | 0.00 |
This paper extends experiments by Beer [Beer, R. D. (1996). Toward the evolution of dynamical neural networks for minimally cognitive behavior. In P. Maes, M. Mataric, J. Meyer, J. Pollack, & S. Wilson (Eds.), From animals to animats 4: Proceedings of the fourth international conference on simulation of adaptive behavior (pp. 421-429). MIT Press; Beer, R. D. (2003). The dynamics of active categorical perception in an evolved model agent (with commentary and response). Adaptive Behavior, 11 (4), 209-243] with an evolved, dynamical agent to further explore the question of representation in cognitive systems. Beer's environmentally-situated visual agent was controlled by a continuous-time recurrent neural network, and evolved to perform a categorical perception task, discriminating circles from diamonds. Despite the agent's high levels of discrimination performance, Beer found no evidence of internal representation in the best-evolved agent's nervous system. Here we examine the generality of this result. We evolved an agent for shape discrimination, and performed extensive behavioral analyses to test for representation. In this case we find that agents developed to discriminate equal-width shapes exhibit what Clark [Clark, A. (1997). The dynamical challenge. Cognitive Science, 21 (4), 461-481] calls ''weak-substantive representation''. The agent had internal configurations that (1) were understandably related to the object in the environment, and (2) were functionally used in a task relevant way when the target was not visible to the agent.