Reinforcement Learning of Dimensional Attention for Categorization
Joshua L. Philips, David C. Noelle · eScholarship (California Digital Library) · 2004
The ability to selectively focus attention on stimulus dimensions appears to play an important role in human category learning. This insight is embodied by learned dimensional attention weights in the ALCOVE model (Kruschke, 1992). The success of this psychological model suggests its use as a foundation for efforts to understand the neural basis of category learning. One obstacle to such an effort is ALCOVE’s use of the biologically implausible backpropagation of error algorithm to adapt dimensional attention weights. This obstacle may be overcome by replacing this attention mechanism with one grounded in the reinforcement learning processes of the brain’s dopamine system. In this paper, such a biologically-based mechanism for dimensional attention is proposed, and the fit of this mechanism to human performance is shown to be comparable to that of ALCOVE.