Parametrization of connectionist models

People involved in the project

Overview

Connectionist models are abstract networks design to model high-level information processing in the brain during a given task. They have been successfully used to explain many effects observed in psychological and imaging experiments.

The behavior of connectionist models is controlled by number of parameters (e.g. weights of connections between nodes, level of noise, nodes' gains, etc.). In order to obtain behavior of the model consistent with the behavior of subjects in an experiment, the values of these parameters have to be set properly.

Below you can download a Matlab script that attempts to find parameters for any connectioinist model that produce specified behavior. The connectionist model needs to be implemented in Matlab, or if it is implemented in any other language, a Matlab script must be written such that the model can be executed by calling a Matlab function.

Click here to download Matlab scripts for parametrization of connectionist models

In order to find out how to use the above script, download and uncompress the scripts, and then type in Matlab:

help fitparam

The above link also includes a very simple example of a connectionist model. To see an example how to use the parametrization routine for this simple model, type in Matlab:

fitparam('example');

The details of the parameterization procedure can be found in file: report.pdf, which is also included among files downloadable from the above link.

The above link also includes Subplex optimization algorithm (simulation have shown that Subplex is most suitable for this optimization problem). The algorithm was developed by Tom Rowan (Intel) and implemented in Matlab by Bruce Lowekamp (College of William and Mary).

Back to Rafal Bogacz Homepage