Efficient Inference in First Passage Time Models

Area of Science:

• Computational cognitive neuroscience
• Mathematical modeling
• Statistical inference

Background:

• First passage time models are crucial for analyzing random processes across scientific disciplines.
• Generalized drift diffusion models (GDDMs) are vital in cognitive neuroscience for understanding decision-making by modeling latent psychological processes.
• Current methods for computing GDDM likelihoods are inefficient when drift rates vary dynamically within trials.

Purpose of the Study:

• To propose a novel, fast, and flexible algorithm for computing the likelihood function of GDDMs.
• To address limitations of existing methods in scenarios with time-varying drift rates.
• To enable more efficient statistical inference for complex GDDMs.

Main Methods:

• Developed a new algorithm for GDDMs satisfying the Cherkasov condition.
• The method segments trials into discrete stages.
• Fast analytical results are used for stage-wise densities, which are then integrated for trial-wise likelihood computation.

Main Results:

• The proposed algorithm provides accurate likelihood evaluations for statistical inference.
• The method significantly outperforms existing approaches in terms of computational speed.
• Demonstrated effectiveness through numerical examples for GDDMs with dynamic drift rates.

Conclusions:

• The new algorithm offers a substantial improvement for analyzing GDDMs, particularly in complex scenarios.
• This advancement facilitates more efficient and accurate parameter estimation in computational cognitive neuroscience.
• The method enhances the applicability of GDDMs to real-world data with dynamic covariates.