Zachary P. Kilpatrick

This paper introduces an information compass framework to model probabilistic reasoning in biological and artificial systems using information geometry.

The paper shows that groups of animals can detect threats more accurately than individuals by using social discounting, where stillness of neighbors signals safety and flight signals danger.

This study investigates how RNNs represent time and make decisions during delayed tasks, finding multiple dynamical solutions with similar behavior.

This paper develops a Bayesian model of foraging in a patchy environment, showing resource depletion accelerates rate learning but not composition learning.

This study examines how humans use flexible decision rules to terminate evidence accumulation, finding that participants adjust their strategies to improve reward rate under varying conditions.

The paper identifies distinct subpopulations of VTA dopamine neurons based on their glutamate and dopamine co-transmission and shows their differential roles in reward and aversion.

The paper introduces a metastable neural circuit model that robustly encodes certainty in memory by using quantized nonlinearities to support multiple bump amplitudes.

The paper shows that human working memory biases are better explained by experience-dependent learning models than by models without learned biases.

This study shows that BNST GABA neurons are necessary for changes in foraging behavior after live predator threats but not robotic threats.

This study trains a minimalist cerebellar cortex model to learn time-series data, revealing that dense granule cell activity and task-specific pattern separation features underlie temporal learning.

This paper derives a normative decision model that uses adaptive thresholds to maximize reward in changing environments, showing better performance than constant-threshold models.

The paper investigates how working-memory limitations affect perceptual decision-making by examining strategy-dependent degradation of a continuous decision variable over time and capacity.

The paper shows that humans use suboptimal inference strategies in rare-event tasks, inverting the bias-variance trade-off compared to optimal Bayesian strategies.

The paper proposes a hierarchical neural network model that simulates perceptual multistability with four stochastically alternating percepts, consistent with experimental observations.

This paper develops a method using differential Chapman-Kolmogorov equations to analyze dynamic decision-making models and compare them to empirical data using accuracy and divergence metrics.

The paper compares normative and approximate evidence accumulation models on a dynamic clicks task, showing parameter tuning differences and pitfalls in model fitting and interpretation.

The paper develops a drift-diffusion model for sequential decision-making, showing how prior information affects future decisions and optimizing reward rates in correlated trial sequences.

The paper develops a normative model for collective decision making in social networks, showing how agents integrate sensory and social evidence to optimize choices.

The paper develops a probabilistic model of trial-to-trial interference in visual working memory and links it to a recurrent network with dynamic connectivity and short-term facilitation.