動的かつ不確実な環境下での適応的知覚と行動に関する研究

通过最优行为的贝叶斯模型来量化不确定性，研究它们与物理中的随机（热）动力学、信息几何中的辛结构和最优控制/强化学习中的目标导向代理的关系，通过其一致的数学公式。该项目包括不确定性下决策机制的数学分析和计算模型。这种不确定性通过最优行为的贝叶斯模型来量化，研究它们与物理中的随机（热）动力学、信息几何中的辛结构和最优控制/强化学习中的目标导向代理的关系，通过它们一致的数学公式。我们的工作提出了一个在神经科学，认知科学和生物学新兴框架的分析：自由能原理。这个框架是一种流行的描述行为和神经活动的方法，用预测模型来编码有关世界的信息。最近，这个数学框架被提出用来解释个性和生命起源的模型，根据与外部世界的统计分离来定义“个体”（甚至是“代理”）。我们的论文对该框架的数学假设、哲学承诺和神经科学应用进行了广泛的回顾和分析，并指出了目前仍然存在的重大缺陷。

Uncertainty is quantified via Bayesian models of optimal behaviour, studying their relations to stochastic (thermo)dynamics in physics, symplectic structures in information geometry and goal-directed agents in optimal control/reinforcement learning via their consistent mathematical formulation.The project includes mathematical analysis and computational models of decision making mechanisms under uncertainty. This uncertainty is quantified via Bayesian models of optimal behaviour, studying their relations to stochastic (thermo)dynamics in physics, symplectic structures in information geometry and goal-directed agents in optimal control/reinforcement learning via their consistent mathematical formulation.Our work presents an analysis of an emergent framework in neuroscience, cognitive science and biology: the free energy principle. This framework is a popular way of describing behaviour and neural activity in terms of predictive models encoding information about the world. More recently, this mathematical framework has been proposed to account for models of individuality and origins of life, casting the definition of “individual” (or even “agent”) in terms of statistical separations from the outside world. Our paper provides an extensive review and analysis of mathematical assumptions, philosophical commitments and applications to neuroscience of this frameworks, identifying significant flaws that still remain at the moment.