CRCNS: Uncovering neurla circuit mechanisms of category computation and learning

描述（由申请人提供）：本研究将探讨视觉分类的皮层回路机制，即学习将视觉刺激分类为具有相同行为意义的物体组的过程。先前的研究表明，在视觉分类任务中，前额叶皮层（PFC）和外侧顶叶间区（LIP）的单个神经元编码刺激的类别隶属性。在这些发现的基础上，我们将结合生物物理逼真的神经模型和行为猴子的单单元记录，来阐明关于类别学习和基于类别的行为的机制问题。首先，我们将开发一个相互作用的感觉回路和顶叶-前额叶回路的尖峰网络模型，以阐明延迟匹配-类别（DMC）任务（样本和测试刺激的属性是否属于同一类别？）与延迟匹配-样本（DMS）任务（样本和测试的属性是否相同？）的关键神经计算的皮层基础。其次，我们将研究如何通过离散的训练阶段学习类别，从基于身份的匹配样本到在任意类别边界附近的刺激下的精细类别区分。这项研究将使用具有奖励依赖性突触学习的模型、猴子行为评估和猴子在不同训练阶段的单单元记录来完成。第三，我们将逐一研究基于身份的DMS与基于类别的DMC之间的任务转换，以阐明刺激身份和类别的不同神经编码，以及在LIP和pfc中视觉分类中的任务规则表示。这些研究将提供重要的见解，并产生一个计算框架，用于理解大脑如何编码视觉刺激的习得性意义或类别隶属性。智力优势：如果没有对刺激进行分类或分类的能力，就很难感知和理解世界；概念和语言似乎是不可能的。因此，阐明分类的神经机制是我们寻求对高级认知的神经生物学理解的关键一步。虽然我们对大脑如何处理感官属性（如运动方向和方向）了解甚多，但对大脑如何实现更抽象的知识获取（如如何通过学习将属性分组为类别）以及基于类别的行为的计算优势知之甚少。在神经回路水平上对这些问题的机械理解需要协调一致的计算和实验努力。因此，我们提出的研究计划的结果很可能代表了这一领域的重大进步，具有广泛的意义。我们非常有前途的初步计算、行为和神经元研究已经验证了我们的方法，并确保了这个项目的各个方面都有很高的成功可能性。更广泛的影响和教育与研究活动的整合：两个pi都积极参与教学。王博士在耶鲁大学教授跨系神经科学研究生课程和新的物理/工程/生物（PEB）综合研究生课程。弗里德曼博士正在为研究生和本科生准备一门名为“神经数据分析方法”的新研讨会课程。课程和练习将围绕计算和统计分析在他的实验室收集的真实数据在这里提出的实验。Wang博士是国际神经信息学协调机构（INCF）神经网络建模描述标准监督委员会的成员。在他的实验室开发的模型将提供给计算界。扩大代表性不足群体的参与——两所学院在招收和指导代表性不足群体学生方面都有良好的记录。目前，王博士有一名女研究生和一名女博士后（Tatiana Engel博士，她将在他的实验室领导拟议的研究）。在过去的两年里，弗里德曼博士实验室的四名研究生来自代表性不足的群体（一名是非裔美国人，其他是女性）。与公众的接触——两个私人机构都积极开展接触活动。王博士在纽黑文的霍普金斯大学做过关于大脑的讲座；弗里德曼博士参与了芝加哥肯伍德学院公立学校的“科学技术推广和指导计划”、“青年科学家培训计划”和学生科学博览会。我们的工作重点是学习和记忆的大脑机制，这是一个既容易接近又引起公众极大兴趣的话题。对于我们的推广和指导工作，我们将使用在拟议工作中产生的数据来制作大脑如何学习和处理视觉信息的教育演示，这些演示将被外行观众访问。这些演示将用于K-12的课堂演示，也可以在网上获得。

DESCRIPTION (provided by applicant): The proposed research will investigate the cortical circuit mechanisms of visual categorization, the process of learning to classify visual stimuli into groups of objects that are equivalent in terms of their behavioral significance. Previous work revealed that individual neurons in the prefrontal cortex (PFC) and the lateral interparietal (LIP) area encode the category membership of stimuli during visual categorization tasks. Built on these findings, we will combine biophysically-realistic neural modeling and single-unit recording from behaving monkeys, to elucidate the mechanistic questions concerning category learning and category-based behavior. First, we will develop a spiking network model of the reciprocally interacting sensory circuit and parieto-prefrontal circuit, to elucidate the cortical basis of key neural computations underlying a delayed match-to-category (DMC) task (do the attributes of a sample and a test stimulus belong to the same category?) versus delayed match-to-sample (DMS) task (are the attributes of the sample and test identical?). Second, we will examine how categories are learnt through discrete training stages, from identity-based match-to-sample to fine category discrimination with stimuli near an arbitrary category boundary. This will be done using models endowed with reward-dependent synaptic learning, monkey behavioral assessment and single-unit recordings from monkeys at different stages of training. Third, we will examine task switching, on a trial-by-trial basis, between the identity-based DMS versus category-based DMC, to clarify the differential neural coding of stimulus identity and category, as well as task-rule representation in visual categorization, in the LIP and PFC. Together, these studies will shed important insights and yield a computational framework for understanding how the brain encodes the learned significance, or category membership, of visual stimuli. Intellectual Merits: Without the ability to classify or categorize stimuli, it would be difficult to perceive and comprehend the world; concepts and language would seem impossible. Therefore, elucidating the neural mechanisms of categorization is a crucial step in our quest for a neurobiological understanding of higher cognition. While much is known about how the brain processes sensory attributes (such as orientation and direction of motion), much less is known about how the brain achieves more abstract knowledge acquisition such as how attributes are grouped into categories through learning, and what are the computational advantages of category-based behavior. A mechanistic understanding of these issues, at the neural circuit level, necessitates a concerted computational and experimental effort. Thus, the results of our proposed research program are likely to represent a significant advance in this area, with broad implications. Our highly promising preliminary computational, behavioral and neuronal studies have validated our approach, and have ensured that all aspects of this project have a high likelihood of success. Broader Impacts and Integration of Education and Research Activities: Both PIs are actively involved with teaching. Dr. Wang teaches for the Interdepartmental Neuroscience graduate program and for the new Physics/Engineering/Biology (PEB) integrated graduate program at Yale. Dr Freedman is preparing new workshop course called "Methods in neuronal data analysis" to both graduate and undergraduate students. Lessons and exercises will revolve around computational and statistical analysis of real data collected in his laboratory during the experiments proposed here. Dr Wang is a member of the Oversight Committee for Description Standards in Neural Network Modeling, International Neuroinformatics Coordinating Facility (INCF). Models developed in his lab will be made available to the computational community. Broaden Participation of under-represented groups-Both PI have a strong track record of recruiting and mentoring students from under-represented groups. At this time, Dr. Wang has a female graduate student and a female postdoctoral fellow (Dr Tatiana Engel who will spearhead the proposed research in his laboratory). Over the past two years four graduate students in Dr. Freedman's laboratory are from underrepresented groups (one is African American and the others are women). Outreach to general public- Both PIs have been active in outreach. Dr Wang has given lectures on the brain at the Hopkins School in New Haven; Dr Freedman has been involved in the "Science and Technology Outreach and Mentoring Program", "The Young Scientist Training Program", and the student science fair at Kenwood Academy public school, in Chicago. Our work focuses on the brain mechanisms of learning and memory, a topic which is both accessible and of great interest to the general public. For our outreach and mentorship efforts, we will use data generated during the proposed work to produce educational demonstrations of how the brain learns and processes visual information that will be accessible to a lay audience. These demonstrations will be used in K-12 classroom presentations and also available online.