bioRxivJClub
Categorical Bayes Filtering for Computational Phenotyping in Adaptive Learning
bioRxiv · · Preprint · Open access
Chen, Piray
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
The Categorical Bayes Filter (CBF) deterministically infers volatility and stochasticity in a large-scale volatility-stochasticity task, outperforming a particle filter in predicting trial-by-trial empirical learning rates.
- Why it matters: Jointly estimating volatility and stochasticity is hard because they require opposite learning-rate adjustments and lack a closed-form way to separate their effects. Estimator noise in particle-filter fitting can also weaken correlations needed for individual-difference analyses.
- What they did: The CBF replaces the stochastic layer of particle-filter models with a categorical distribution over a quantile grid using differentiable Beta quantile functions, enabling evidence maximization with an exact, deterministic, differentiable marginal likelihood. It was fit in a large-scale volatility-stochasticity task and evaluated against a state-of-the-art particle filter.
- The result: CBF dispersion parameters captured volatility-blind and stochasticity-blind participants, and trial-by-trial ambiguity predicted response times; fitted parameters also predicted psychiatric symptoms in the hypothesized direction. This makes individual-level analyses with continuous-latent particle-filter models practical for cognitive modeling and computational psychiatry.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.
- Open access
Free. We’ll email you when a new paper cites this one or any other paper you follow on JClub — at most one email a day. Every email has a one-tap unsubscribe.