Rafał Bogacz

This paper introduces a model where neurons adjust synaptic strengths to minimize somato-dendritic mismatch, enabling network learning and reproducing experimental observations of plasticity.

This paper models dopaminergic signals linked to habit formation using temporal-difference action learning, showing that action prediction errors better represent striatal dopamine activity than reward prediction errors.

The paper finds evidence of traveling beta waves in the subthalamic nucleus of Parkinson's patients using local field potential recordings and a spherical wave model.

This paper investigates temporal predictive coding (tPC) as a framework for learning temporal dependencies in recurrent neural networks, showing its equivalence to BPTT variants and its ability to handle complex tasks with local plasticity.

This study shows that dithering suppresses half-harmonic neural synchronisation to photic stimulation in humans compared to periodic stimulation.

This study shows dopamine in the dorsal striatum encodes both sensory and reward prediction errors during perceptual decisions, with distinct but overlapping signals.

This study shows that dopamine signals in the striatum reflect long-term learning strategies in mice, with a deep neural network model capturing these dynamics through reinforcement learning.

The paper validates a mathematical model's predictions about neuronal population responses to phase-locked stimulation using ECoG data from Parkinsonian rats.

This paper introduces Monte Carlo predictive coding, a framework that unifies predictive coding and neural sampling to model sensory input distributions using local computation and plasticity.

The paper shows mice adapt perceptual choice history biases to different temporal regularities, supported by a reinforcement learning model and striatal dopamine signals.

This paper develops a theoretical framework to determine optimal brain stimulation strategies for modulating phase-amplitude coupling in neurological disorders.

This paper shows that predictive coding models can detect novelty at different levels of representation hierarchy, linking novelty detection, memory, and learning.

The paper proposes a model linking dopamine signaling to uncertainty-driven exploration in reinforcement learning, showing improved performance in simulations and behavioral data.

This paper introduces a temporal predictive coding model implemented in recurrent networks that can predict future stimuli and produce neuron-like responses without complex computations.

This paper introduces a new credit assignment mechanism called prospective configuration, showing it can enable efficient learning in biological contexts and replicate observed neural behaviors.

This study develops a model to predict cortical gamma oscillation entrainment to DBS in Parkinson's patients and validates it in three subjects.

The paper proposes a temporal difference model that explains how dopamine neurons encode different reward dimensions and adapts behavior based on physiological needs.

This paper proposes predictive coding models that implicitly learn covariance information for associative memory, unifying predictive coding and covariance learning in hippocampal networks.

This study shows that hunger affects risk preferences more when risks are learned through experience than when they are explicitly described.

This paper proposes dithered brain stimulation to selectively entrain specific neuronal rhythms while avoiding harmful sub- and super-harmonic entrainment.

This study investigates how subthalamic nucleus activity modulates action adaptation in Parkinson's patients during a force adaptation task, finding that reduced beta activity correlates with better adaptation.

This paper introduces a new model of learning in the basal ganglia that scales prediction errors by reward uncertainty, showing improved performance in noisy environments and alignment with physiological features.

The paper examines neural correlates of conflict and evidence integration during sequential decision making, finding increased cortico-subthalamic coherence and prolonged subthalamic beta oscillations during conflict.

The paper introduces an extension to temporal difference learning that enables instantaneous adaptation to changing reward functions without retraining.

This study shows that hunger enhances reinforcement-driven decision-making but not planned actions, suggesting a bias toward primitive learning when hungry.

This paper develops an algorithm for optimal closed-loop deep brain stimulation using multi-contact electrodes to maximize desynchronization of brain activity.

This paper shows that average beta burst duration profiles reflect changes in the dynamics of beta oscillations between ON and OFF medication states in Parkinson's disease, linking them to motor impairment.

This paper models dopamine's role in learning and action inference by linking prediction errors to reward and habit formation in the basal ganglia.

The paper identifies signatures of the hyperdirect pathway and its role in pathological network disruption in Parkinson's disease through multimodal analysis.