Samuel Joseph Gershman

This paper proposes an online hierarchical Bayesian inference model to explain human motion perception and demonstrates its ability to qualitatively and quantitatively explain various motion stimuli.

The paper provides neural evidence for the successor representation in choice evaluation by linking behavioral errors to predictive representations in sensory cortex.

The paper uses a Bayesian model to show how dopaminergic control influences impulsivity and risk-seeking in intertemporal choice and probability discounting tasks.

The paper investigates how causal beliefs influence reinforcement learning by analyzing fMRI data from human participants during a task manipulating causal structure beliefs.

This paper introduces the concept of rational inattention to reconcile two theories of tonic dopamine's role in reinforcement learning and interval timing, showing how DA balances cognitive costs and reward availability.

This study shows human motion perception aligns with Bayesian structural inference in psychophysics experiments, matching performance and error patterns.

The paper investigates how inferred controllability modulates Pavlovian and instrumental action selection using neural data from a Go/NoGo task.

This study investigates human multi-task reinforcement learning by comparing participant behavior to algorithms in a two-step decision-making task with changing rewards.

This study examines the reward-complexity trade-off in schizophrenia patients and healthy controls, finding similar empirical trade-off curves but differing policy complexity and bias.

This study demonstrates that ramping dopamine signals correspond to reward prediction errors rather than state values, unifying rapid phasic and slow ramping signals as derivative-like computations.

The paper investigates how humans use hierarchical motion structure in visual perception through controlled stimuli and experiments.

This paper formalizes hippocampal remapping as hidden state inference, resolving ambiguities about context changes and remapping classification.

The paper identifies distinct neural correlates of uncertainty in the brain that underlie different exploration strategies during decision-making.

This paper introduces a Bayesian model of hierarchy discovery that explains how humans organize environments into hierarchical structures for planning, supported by simulations and behavioral experiments.

The paper introduces a neuro-symbolic model called SEM that organizes continuous experiences into discrete events and generalizes memory, showing human-like event segmentation in video data.

The paper investigates how controllability affects the balance between Pavlovian and instrumental action selection, finding that Pavlovian bias is stronger in low control contexts.

The paper shows that groups of dopamine neurons in rats and human midbrain signal specific sensory prediction errors through ensemble firing patterns.

This paper proposes a new theory of dopamine function that extends prediction error to include sensory predictions, bridging model-based and model-free reinforcement learning.

The paper shows that dopamine-related genes influence both directed and random exploration strategies in a task involving reward maximization.

This study uses fMRI to investigate neural correlates of structure learning and associative learning, showing distinct brain regions and predictive behavioral outcomes.

The paper identifies pure neural correlates of exploration and exploitation in the human brain using an observe-or-bet task.

This paper proposes that the hippocampus uses a predictive representation for spatial navigation, challenging the traditional cognitive map metaphor and suggesting grid cells support hierarchical planning.

The paper explains dopamine responses through Bayesian reinforcement learning, accounting for deviations from the reward prediction error hypothesis.

This paper proposes a framework linking model-based reinforcement learning to model-free mechanisms using successor representations and introduces algorithms that mitigate their limitations.

This study investigates the successor representation (SR) in human reinforcement learning, showing it balances flexibility and efficiency by being insensitive to task sequence changes but responsive to reward changes.

This paper reviews how computational theories can inform statistical models of neural data by providing constrained priors, enabling data-driven evaluation of theories.

The paper provides behavioral evidence for the penumbra of learning in humans and shows it can be explained by a computational theory of learning.