权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Modeling and mapping multiple computational processes in human reinforcement learning

人类强化学习中的多个计算过程的建模和映射

基本信息

批准号：
9761233
负责人：
Samuel David McDougle
金额：
$ 6.09万
依托单位：
UNIVERSITY OF CALIFORNIA BERKELEY
依托单位国家：
美国
项目类别：
财政年份：
2019
资助国家：
美国
起止时间：
2019-03-01 至 2022-02-28
项目状态：
已结题

来源：
https://reporter.nih.gov/project-details/9761233
关键词：
Addictive Behavior Address Artificial Intelligence Attenuated Automobile Driving Basal Ganglia Behavior Behavioral Biological Biological Assay Brain Categories Choice Behavior Clinical Protocols Cognitive Computer Simulation Data Dopamine Environment Event Exposure to Feedback Functional Magnetic Resonance Imaging Future Habits Human Individual Knowledge Learning Learning Module Mediating Mental disorders Methods Mind Modeling Nervous system structure Neurosciences Operant Conditioning Parietal Parkinson Disease Participant Phase Play Process Psychological reinforcement Psychology Psychophysics Reaction Time Research Response to stimulus physiology Rewards Role Scanning Schizophrenia Series Short-Term Memory Signal Transduction Study models Support System System Techniques Testing Time Variant Weight Work addiction base clinical application clinical development cognitive process cost design experimental study flexibility human model improved learning algorithm neural circuit neuroimaging neuromechanism neurophysiology novel psychologic recruit relating to nervous system theories visual motor visual stimulus

项目摘要

Project Summary Learning from rewards is one of the fundamental roles of the nervous system, allowing for beneficial behaviors to be repeated, and detrimental behaviors to be avoided. It has recently become clear that when humans learn from rewards in the environment, they rely on multiple neural systems that work in tandem. What are the psychological and biological constraints of these systems, and how do they interact during learning? The proposed experiments are designed to answer these questions by developing precise computational models of human instrumental learning, as well as investigating the neural dynamics of, and interactions between, individual learning processes. Aim 1 will focus on isolating individual learning processes, further developing a novel model of human instrumental learning that highlights contributions to learning from both a flexible executive working memory module and an incremental reinforcement learning module. Behavioral experiments and computational modeling will be used to better characterize these two learning processes, with a focus on how they interact instantaneously and over time. Aim 2 will use a combination of brain stimulation and neuroimaging to better characterize the neural systems supporting each learning process, as well as the putative interactions between these neural circuits. These results will constrain our understanding of the neural mechanisms that drive human instrumental learning. The knowledge gained by this project will provide a vital framework for clinical applications, for instance, in understanding and treating working memory related learning deficits in schizophrenia, and reinforcement learning deficits in Parkinson's disease. Critically, a more precise model of individual learning processes could guide the development of clinical protocols that leverage intact learning systems when other learning systems are compromised. Finally, an enhanced understanding of human reward-based learning could improve theories of habit formation, which may further inform psychological and neurophysiological models of addiction.

项目概要从奖励中学习是神经系统的基本作用之一，它允许有益的行为重复，并避免有害行为。最近很明显，当人类学习时从环境中的奖励来看，它们依赖于协同工作的多个神经系统。有哪些这些系统的心理和生物限制，以及它们在学习过程中如何相互作用？这所提出的实验旨在通过开发精确的计算模型来回答这些问题人类器乐学习，以及研究神经动力学及其之间的相互作用，个人学习过程。目标 1 将侧重于隔离个人学习过程，进一步开发人类器乐学习的新颖模型，强调了灵活的学习方式对学习的贡献执行工作记忆模块和增量强化学习模块。行为实验计算模型将用于更好地描述这两个学习过程，重点是他们如何即时和随着时间的推移相互作用。目标 2 将结合使用大脑刺激和神经影像学可以更好地表征支持每个学习过程的神经系统，以及这些神经回路之间的假定相互作用。这些结果将限制我们对神经网络的理解驱动人类乐器学习的机制。通过该项目获得的知识将为临床应用框架，例如理解和治疗与工作记忆相关的学习精神分裂症的缺陷和帕金森病的强化学习缺陷。关键的是，更精确的个人学习过程的模型可以指导利用完整的临床方案的开发当其他学习系统受到损害时，学习系统就会受到影响。最后加深了对人类基于奖励的学习可以改善习惯形成的理论，这可能会进一步提供信息成瘾的心理和神经生理学模型。