Computational frameworks for decision-making: From bayesian inference to reinforcement learning models

The ability to make adaptive decisions in uncertain environments is a fundamental characteristic of biological intelligence. Historically, computational neuroscience has utilized two dominant paradigms to explain this behavior: Bayesian Inference, which describes how organisms maintain and update internal representations of the world, and Reinforcement Learning (RL), which details how agents learn optimal actions through trial and error. While Bayesian models excel at capturing the integration of prior knowledge with sensory evidence, RL frameworks provide a robust account of value-based learning and the role of neuromodulators like dopamine. Recently, the boundary between these two fields has blurred, leading to "Bayesian Reinforcement Learning" and other hybrid architectures. Despite this progress, understanding how the brain physically implements these complex algorithms across different temporal and structural scales remains a cornerstone challenge in modern neuroscience.The primary objective of this Research Topic is to bridge the gap between high-level algorithmic theories of decision-making and their biological underpinnings. While abstract models have reached a high level of sophistication, we still lack a comprehensive understanding of how neural circuits realize these computations, particularly in dynamic, non-stationary environments. This collection aims to address several critical questions: How does the brain represent uncertainty within RL frameworks? What are the common computational principles that link probabilistic inference with reward-based learning?By soliciting interdisciplinary contributions, we aim to synthesize current knowledge and foster the development of "next-generation" models that are both biologically plausible and computationally efficient. We seek to provide a platform for research that explores the synergy between Bayesian and RL approaches, providing new insights into how biological systems achieve near-optimal decision-making despite inherent noise and limited resources.We welcome articles that utilize computational modeling, behavioral experiments, and neurophysiological data. We are particularly interested in work that integrates multiple levels of analysis, from synaptic plasticity to large-scale network dynamics.Specific themes of interest include, but are not limited to:- Hybrid Frameworks: Integration of Bayesian priors into Reinforcement Learning architectures.- Neural Implementation: Mechanisms of how cortical and subcortical circuits represent probability distributions and value functions.- Uncertainty and Exploration: Computational accounts of the exploration-exploitation trade-off in complex environments.- Hierarchical Decision-Making: How models scale from simple binary choices to multi-step, goal-directed planning.- Clinical Applications: Computational psychiatry approaches using these frameworks to explain maladaptive decision-making in disorders such as addiction or anxiety.- Comparative Analysis: Performance and structural comparisons between biological decision-making and artificial intelligence agents. Topic editors: Hongji Xu, Alfredo Garcia, Paul Miller, Kevin Marche Source: https://www.frontiersin.org/research-topics/80546/computational-frameworks-for-decision-making-from-bayesian-inference-to-reinforcement-learning-models