New Method Designs Optimal Experiments for Cognitive Model Inference

Researchers have developed a novel approach to optimize the design of cognitive experiments, aiming to improve the accuracy of inferring underlying cognitive processes from observed behavior. The method, based on Bayesian Experimental Design (BED), treats the experimental environment itself as a variable to be optimized, moving beyond traditional approaches where environments are fixed. The study establishes a Monte Carlo BED benchmark and introduces an amortized BED framework, significantly reducing computational costs while maintaining accuracy in identifying optimal environments. Experiments using the Mouselab-MDP paradigm revealed that different cognitive inference objectives require distinct experimental designs, indicating trade-offs between maximizing information gain, ensuring posterior recoverability, and achieving information efficiency. This framework provides a principled way to construct more effective cognitive experiments, leading to more robust and informative parameter inference for computational cognitive models.