An Overcomplete Approach to Fitting Drift-Diffusion Decision Models to Trial-By-Trial Data.

Area of Science:

• Computational Neuroscience
• Decision Science
• Cognitive Psychology

Background:

• Drift-diffusion models (DDMs) are standard for explaining decision outcomes and reaction times (RT).
• Current DDM parameter estimation methods face limitations with complex model variants and computational feasibility.
• Existing approaches struggle with drift rates varying over trials or collapsing bounds.

Purpose of the Study:

• To propose a fast and efficient parameter estimation method for DDMs.
• To overcome computational bottlenecks in standard DDM fitting.
• To enable reliable parameter estimation for complex DDM variants.

Main Methods:

• Developed a novel approach based on fitting a "self-consistency" equation for RTs under DDMs.
• Utilized a variational Bayesian system identification scheme to handle model "overcompleteness" arising from estimating neural noise variables.
• Employed numerical simulations to validate the method's performance.

Main Results:

• The proposed method matches current approaches for simple DDMs.
• It outperforms existing methods for more complex DDM variants, including those with collapsing bounds.
• Demonstrated the approach's added value in analyzing value-based decision-making experiments.

Conclusions:

• The novel self-consistency equation approach offers a computationally efficient and accurate method for DDM parameter estimation.
• This method uniquely provides estimates of neural noise perturbations, valuable for neural data analysis.
• The approach enhances the analysis of decision-making processes, particularly for complex scenarios.