Adaptive design optimization: a mutual information-based approach to model discrimination in cognitive science

Area of Science:

• Cognitive Science
• Computational Statistics
• Experimental Design

Background:

• Accurate statistical model discrimination is crucial for cognitive science research.
• Designing maximally informative experiments is key to efficient model inference.
• Analytical solutions for optimal designs are often intractable for nonlinear models.

Purpose of the Study:

• To develop a numerical method for optimizing experimental designs in cognitive science.
• To enable efficient discrimination among competing statistical models, including nonlinear ones.
• To identify experimental designs that allow model inference in the fewest possible steps.

Main Methods:

• Employed a Bayesian computational trick to recast adaptive design optimization.
• Formulated the problem as a probability density simulation.
• Utilized a utility function based on mutual information.

Main Results:

• The optimal experimental design corresponds to the mode of the simulated probability density.
• Provided three intuitive interpretations of the mutual information utility function.
• Demonstrated the method's efficacy with an example application in memory retention experiments.

Conclusions:

• Numerical Bayesian methods offer a viable solution for optimizing experimental designs with complex models.
• The proposed approach facilitates efficient model discrimination in cognitive science.
• This work provides a practical tool for designing more informative cognitive experiments.