Computational principles of dual contribution of dopamine in reinforcement learning and action selection

Authors: Roberto Sautto, Nicolas Cuperlier, Thanos Manos, Marwen Belkaid

Presentation type: Poster at SNUFA 2026 online workshop (4-5 Nov 2026)

Abstract

When making decisions in volatile environments biological agents are able to take into account the statistics of action-outcome contingencies and use them to select appropriate behaviour. In this context, Dopamine has been widely studied as a reinforcement signal for synaptic plasticity, but it has also been shown to affect action selection and contribute to the regulation of the exploration-exploitation trade-off. How the dopaminergic system reconciles these two functions is not fully understood yet.

In this work, we present a spiking neural network model of Dopamine-driven reinforcement learning and decision-making.

First, we compare two models of Dopamine-driven learning which compute reward prediction errors respectively using direct and disinhibition-based dopaminergic control. We show that both mechanisms produce expected firing regimes, but that the disinhibition model is better able to learn about reward statistics.

Then, we fit this model against behavioural data from mice performing a 3-armed bandit task. We examine how the dopaminergic signal emerging from the interaction of neural populations in the model enables it to both acquire expected values from the task, as well as affect action selection and the exploration-exploitation tradeoff. Thus, this work aims to uncover the computational principles underlying the dual role of Dopamine as a reinforcer and modulator, and has implications for the study of decision-making brain circuits.

Are you the author and do you want to add some related links to this page? If so, check out our guide to adding related links and videos.