Computational optimization of associative learning experiments.

Area of Science:

• Computational biology
• Theoretical biology
• Behavioral science

Background:

• Increasingly complex computational models are crucial for accurate biological system accounts.
• Designing optimal experiments for complex models is challenging due to their flexibility.
• Associative learning research faces difficulties in manually designing informative experiments for complex models.

Purpose of the Study:

• To propose a novel method for computational optimization of associative learning experiments.
• To address the challenge of designing efficient experiments for complex computational models.
• To improve parameter estimation and model selection in scientific research.

Main Methods:

• Formalizing associative learning experiments with tunable design variables for tractability.
• Combining simulation-based Bayesian experimental design with Bayesian optimization.
• Validating the method through extensive simulations for parameter estimation and model selection.

Main Results:

• Computationally optimized designs significantly outperform manual designs, even with limited prior information.
• The proposed method offers a flexible approach to tuning experimental design variables.
• Optimized designs enhance the expected utility of studies, potentially improving reproducibility.

Conclusions:

• Computational optimization of experiments is a powerful tool for advancing scientific research, particularly in associative learning.
• This approach can enhance reproducibility and encourage practices like pre-registration.
• The method holds promise for both basic and translational research, including computational psychiatry.