Active Inference in Hierarchical Brain Networks: Mechanisms, Functions, Modulation.

大脑的一项非凡壮举是它能够将数十亿个神经细胞的活动整合成一个统一的自我。然而，所涉及的机制的性质基本上是未知的。此外，这种惊人的能力是脆弱的，当我们面对大脑和精神的痛苦时，我们仍然一无所知。因此，我的研究计划的首要目标是推进我们对大脑系统功能整合的机械性理解。我的研究理念是通过大脑回路进行主动推理。通俗地说，我认为大脑是一台预测机器，它主动预测自己的输入，评估与实际外部事件的差异，并相应地更新内部语境表征，以适应行为和学习。从机制上讲，我假设这种恒定的、经常性的流量是由大脑网络中相互依赖的节律波动实现的。我的方法是多尺度电生理学(e-phys；从细胞到脑电/脑磁图的全脑成像)。我还使用了一系列大脑刺激技术(光遗传学、经颅刺激、实时闭环感觉刺激)来最好地捕捉动物模型和人类志愿者大脑网络的快速动态，并对其进行因果修改。我还用自然刺激序列训练人工代理，以在大脑网络中推导出主动推理的代理，并产生神经生理活动的编码模型。我的探索研究计划结构如下：？目标1(模型和工具)将提供受生物学启发的具有耦合振荡器的振荡大脑网络的可测试的概念模型，以阐明主动推理的机制。用于计算机密集型数据分析的免费、开源软件套件将能够对大量密集的神经生理学数据进行建模和处理；目标2(功能)将提供大脑主动推理的经验证据，集中于听觉工作记忆和对自然输入的感知；目标3(调制)将提出新的循证方法，用于靶向神经调节，包括生物启发的脑训练策略，以确定脑功能和行为中多节律组合波动的因果作用。我的计划建立在我的实验室在研究大脑系统的神经动力学方面的优势之上。它还将通过我已建立的合作者网络，为我的研究提供新的基本方向和翻译方向。我将进一步致力于与一个行业合作伙伴将研究成果切实转移到可穿戴、低成本的设备上。总体而言，预期结果将显著提高对大脑功能综合机制的基本知识。我预计，公布的原则和提供的工具将为研究社区服务，以回答广泛的系统神经科学问题。

A remarkable feat of the brain is its capacity to integrate the activity of billions of nervous cells into a unified self. Yet, the nature of the mechanisms involved are essentially unknown. Further, this astonishing competence is vulnerable, and we remain clueless when confronted with the afflictions of the brain and the mind. The overarching goal of my research plan is therefore advance our mechanistic comprehension of functional integration in brain systems. My research concept is that of active inference by brain circuits. In lay terms, I see the brain as a predictive machine that actively forecasts its inputs, evaluates discrepancies with actual external events, and updates internal contextual representations accordingly for behavioral adaptation and learning. Mechanistically, I posit that this constant, recurrent flux is implemented by interdependent rhythmic fluctuations in brain networks. My methods are those of multiscale electrophysiology (e-phys; from cells to whole-brain imaging with EEG/MEG). I also use a range of brain stimulation techniques (optogenetics, transcranial stimulation, real-time closed-loop sensory stimulation) to best capture and causally modify the rapid dynamics of brain networks in animal models and human volunteers. I also train artificial agents with natural stimulus sequences to derive proxies of active inference in brain networks and produce encoding models of neurophysiological activity. My Discovery research plan is structured as follows: ? Aim 1 (Models & Tools) will deliver testable, conceptual models of oscillatory brain networks with coupled oscillators inspired by biology, to clarify the mechanisms of active inference. A free, open-source software suite for computer-intensive data analytics will enable the modeling and processing of large and dense neurophysiological data volumes; ? Aim 2 (Functions) will provide empirical evidence of brain active inference with a cohesive focus on auditory working memory and perception of naturalistic inputs; ? Aim 3 (Modulation) will propose new evidence-based approaches for targeted neuromodulation, including biologically-inspired brain training strategies, to establish the causal role of polyrhythmic ensemble fluctuations in brain functions and behavior. My program builds on my lab's strengths in the study of neural dynamics of brain systems. It will also support new fundamental and translational directions in my research, with my established network of collaborators. I will further aim at practical transfers of research outcomes to wearable, low-cost devices with an industrial partner. Overall, the expected outcomes will significantly advance basic knowledge of integrative mechanisms of brain functions. I anticipate the principles unveiled and tools deliver will serve the research community to answer a broad range of systems neuroscience questions.