Model on automaticity of rule-guided behaviors

A biologically-detailed computational model of how rule-guided behaviors become automatic.

• How the model works
• Understanding the code

How the model works

The model is run by specifying a model configuration and passing it as an argument (along with other parameters, if desired) to the function automaticityModel (found in automaticityModel.m).

The automaticityModel runs through logic which is roughly segmented by the following stages:

• Pre-processing - The model initializes variables and objects.
  • Initialize configuration-specific and configuration-agnostic variables.
  • Create neuron objects.
• Processing - The model simulates a specified number of trials.
  • Set a visual stimulus and simulate its effects on the prefrontal cortex (PFC) and premotor cortex (PMC) neurons.
  • Simulate the effects of the visual stimulus on the set of neurons in the model over a discrete time interval.
  • Record data for the trial on which neurons exhibited the strongest reactions.
  • If applicable, apply Hebbian learning to modify local weights for a neuron.
• Post-processing - The model displays results or does other processing (e.g., calculating global parameter optimization values).

Model configurations

A model configuration affects how the model behaves. This includes (but is not limited to):

• Affecting how many trials the model runs through.
• Determining for which trials learning is active.
• Deciding if the COVIS model of rule learning is enabled.
• Deciding if the FROST model of working memory maintenance is used.

Configurations are represented as classes. Configurations inherit from the class ModelConfig, which contains logic shared between all configurations. These classes can be found in internal/config.

The following configurations are available, along with their in-code class name.

• Electrophysiology/ModelConfigElectro - Wallis, Jonathan D., and Earl K. Miller. “From Rule to Response: Neuronal Processes in the Premotor and Prefrontal Cortex.” Journal of Neurophysiology, vol. 90, no. 3, 7 May 2003, pp. 1790–1806., doi:10.1152/jn.00086.2003.
• Dual-task interference/ModelConfigDualTask - Zeithamova, Dagmar, and W. Todd Maddox. “Dual-Task Interference in Perceptual Category Learning.” Memory & Cognition, vol. 34, no. 2, 2006, pp. 387–398., doi:10.3758/bf03193416.
• Button-switch interference/ModelConfigButtonSwitch - Helie, S., et al. “Evidence for Cortical Automaticity in Rule-Based Categorization.” Journal of Neuroscience, vol. 30, no. 42, 2010, pp. 14225–14234., doi:10.1523/jneurosci.2393-10.2010. Referenced when working with FMRI data and optimization.

Sample usage

The following code runs the model using the button-switch interference configuration (see example.m).

disp("Loading all functions and classes into path ...");
init;
disp("Functions and classes loaded onto path.");

disp("Generating performant code ...");
compile;
disp("Performant code generated.");

disp("Running model ...");
optional_parms = struct( ...
	'VIS_INPUT_FROM_PARM', 0, ...
	'visualinput', zeros(2) ...
);
[config, neurons] = automaticityModel_mex( ...
    getmodelparams(ModelConfigButtonSwitch()), ...
    optional_parms ...
);

Understanding the code

Toolbox requirements

Required to run model

• MATLAB R2020A
• MATLAB 9.1
• Curve Fitting Toolbox
• MATLAB Coder
• MATLAB Compiler

Used in other parts of the project

• Global Optimization Toolbox
• Optimization Toolbox
• Parallel Computing Toolbox
• Statistics and Machine Learning Toolbox

Project structure

• /data - Matrix files and related code, categorized by configuration.
• /external - External library dependencies.
• /internal - Code internal to this project. Most of the implementation can be found here.
• /output - Output of each run (state of the configuration and neurons) is automatically saved to this directory.
• .gitignore - Version control file that specifies which files/directories should be ignored when updating this repository.
• automaticityModel.m - The function which runs an automaticity configuration through a series of trials. Can be thought of as the "main" function for this project.
• compile.m - Generates performant code for automaticityModel.m.
• example.m - Example usage of the model.
• init.m - Utility script to recursively load all folders from the current working directory. Useful for ensuring all classes/functions are in memory before doing work with the model.

Code-generation and performance

Unfortunately, running automaticityModel by itself can be rather slow. In order to achieve acceptable performance, it is necessary to leverage code generation to compile automaticityModel into more performant MEX code.

This has placed certain constraints on how the code is written:

• Handle classes cannot be used, so all methods that result in state change (e.g., configuration, neurons) must still assign their output object to their original variable.
• Function argument validation is not supported.