A probabilistic model for noisy decision making in rodent brains

Official URL: https://dx.doi.org/10.1109/SIU71813.2026.11636929

Decision-making provides a foundational framework for studying cognition, yet it is an inherently noisy process. Behavioral variability arises from multiple sources, including sensory noise, decision biases, and motor errors, which are often conflated in existing computational models. We introduce a probabilistic framework that explicitly separates these components in a two-alternative forced-choice interval discrimination task. Subjective time is modeled as a Gaussian random variable with variance proportional to interval duration, consistent with scalar variability. The resulting noisy percept is combined with a Win-Stay, Lose-Switch history feature and a static bias to form a decision variable that governs probabilistic action selection with an explicit lapse process. Model parameters are estimated using gradient-based maximum likelihood optimization enabled by the reparameterization trick. The model is fitted to behavioral data from mice and provides a quantitative framework for dissociating sensory uncertainty, history dependence, and motor bias in temporal decision-making. Results demonstrate that the model captures psychometric sensitivity and behavioral performance with high fidelity, showing no statistically significant deviation from empirical behavior. From an engineering perspective, this framework offers a robust method for designing biologically plausible decision-making systems that function under uncertainty and provides a computational tool for quantifying phenotypic deficits in neurological research.