Research Seminar
The Ising Decision Maker: a thermodynamical approach to decision RT modelingStijn VerdonckKU Leuven | |
| Abstract: | The elementary decision making process is among the most intensively investigated concepts in the current field of psychology. The traditional modeling approach of defining abstract information accumulators on a macroscopic level (with the Ratcliff diffusion model as a prime example) has been successful in describing a number of RT phenomena. Mostly formulated as linear dynamical systems, these models are relatively easy to manipulate and parametrize, and therefore a popular choice for modeling RT data. The computational properties of any biological system however, emerge from a microscopic, in this case neural, level. In recent years, a plausible integrate and fire network has been proposed modeling the 2AFC process bottom up (Wong & Wang, 2006), but its computational complexity renders it less attractive for the psychological researcher desiring a efficiently estimable model. Furthermore, a number of issues are left unaddressed (e.g., fast error RTs, SAT). To alleviate these problems, we propose an abstract version of this network based on the Ising model: a collection of N pairwise interacting neuron-bits. While retaining the basic qualities of a noisy multiple attractor network, it is not concerned with the concrete mechanisms of elementary neuronal interaction. As a thermodynamical system, the N-dimensional microscopic description can be reduced to a two-dimensional macroscopic model, closely connected to a set of two non-linear diffusion equations. It will be shown that the Ising Decision Maker is able to reproduce many empirical phenomena, among which fast and slow errors, SAT and Weber's law. |
| Date: | Tue Nov 8, 12:00 pm - 1:00 pm |
| Place: | room 01.07 (Department of Psychology, Tiensestraat 102, 3000 Leuven) |