CRCNS Research Proposal: Network models of cortical and subcortical interactions for dynamical control of decision making

大脑中的决策是由一个相互作用的大脑区域网络共同做出的，每个区域在决策过程中扮演不同的角色。这些分布式网络是灵活的，能够有效地响应不断变化的环境，同时也非常强大，即使在部分中断或损坏时也能够保持功能。该项目旨在了解这些大脑区域在决策过程中如何相互作用，以及这些相互作用如何赋予它们显着的灵活性和鲁棒性。为了做到这一点，研究团队将联合收割机结合尖端技术，记录和修改神经元的活动，而大鼠作出决定，与机器学习技术建模生成的数据。该项目将提高对更普遍地支持决策和认知的大脑系统的理解，同时也为下一代大脑启发的人工智能系统提供关键的洞察力，该提案结合了清醒时的大规模多区域记录行为大鼠和数据驱动的递归神经网络建模，通过其对相互作用的皮层下区域的影响来研究关联皮层在决策过程中的作用。该项目的第一部分将使用Neuropixels记录沿着多区域递归神经网络建模，以确定关联皮层是否以及如何在决策过程中控制相互关联的皮层下动力学中发挥作用。记录将针对两个区域的关联皮层和两个皮层下区域，而大鼠执行决策任务，需要灵活整合嘈杂的感官信息随着时间的推移。网络建模将用于消除每个大脑区域在决策形成过程中塑造神经动力学的特定角色的竞争假设。该项目的第二部分将通过光遗传学将联合收割机类似的网络建模与大脑活动的实验扰动相结合，以确定神经决策鲁棒性的基础机制。研究人员将确定结构鲁棒性的机制，这产生于分布式网络本身的架构，以及动态鲁棒性的机制，这涉及到对扰动的主动补偿。通过这种方式，该项目将协同机器学习和系统神经科学工具领域最近令人兴奋的进展，创建一个紧密的实验-理论循环，以解决具有广泛跨学科重要性的问题，该奖项反映了NSF的法定使命，并被认为是值得通过使用基金会的智力价值和更广泛的影响审查标准。

Decisions in the brain are collectively made by a network of interacting brain regions that each play different roles in the decision-making process. These distributed networks are flexible, able to respond effectively to changing circumstances, while also highly robust, and able to preserve functionality even when partially disrupted or damaged. This project seeks to understand how these brain regions interact with each other during decision making and how these interactions confer their remarkable flexibility and robustness. To do this, the research team will combine cutting edge technologies for recording and modifying the activity of neurons while rats make decisions, with machine learning techniques for modeling the data generated. The project will improve understanding of brain systems that support decision making and cognition more generally, while also providing critical insight for the next generation of brain-inspired artificial intelligence systems.The proposal combines large-scale multi-region recordings in awake, behaving rats and data-driven recurrent neural network modeling to investigate the role of association cortex during decision-making through its impact on interacting subcortical areas. The first part of the project will use Neuropixels recordings along with multi-area recurrent neural network modeling to identify whether and how association cortex plays a role in controlling interconnected subcortical dynamics during decision making. The recordings will be targeted to two areas of association cortex and two subcortical regions while rats perform decision tasks that require flexible integration of noisy sensory information over time. The network modeling will be used to disambiguate competing hypotheses for the specific roles of each brain region in shaping neural dynamics during decision formation. The second part of the project will combine similar network modeling with experimental perturbation of brain activity through optogenetics to identify mechanisms that underlie the robustness of neural decision making. The investigators will identify mechanisms of structural robustness, which arise from the architecture of the distributed network itself, and mechanisms of dynamic robustness, which involve active compensation for perturbation. In this manner, the project will synergize recent exciting advances in both machine learning and tools for systems neuroscience to create a tight experiment-theory loop to address questions with broad interdisciplinary importance, with implications at the core of developing principled treatments for cognitive disorders.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.