CRCNS US-German Research Proposal: Efficient representations of social knowledge structures for learning from a computational, neural and psychiatric perspective (RepSocKnow)

项目描述 美国-德国计算神经科学合作研究提案： 从计算神经网络学习的社会知识结构的有效表示 和精神病学观点（RepSocKnow） 美方主要研究者：Prof. Gabriela Rosenblau，Ph.D.，助理教授，心理学系，乔治 地址：2115 G Street NW，华盛顿，DC 20052 德方主要研究者：Christoph W.科恩博士，助理教授& PI Emmy-Noether集团，科 社会神经科学，普通精神病学系，海德堡大学，Vossstraße 4，69115 德国海德堡 顾问1：Daniela Schiller教授，博士，精神病学系副教授 弗里德曼脑研究所，西奈山伊坎医学院，1470麦迪逊 Ave，纽约10029，NY 顾问2：Jan Gläscher博士，PI伯恩斯坦研究小组，系统神经科学研究所， Hamburg-Eppendorf大学医学中心，Martinistrasse 52，20246 Hamburg，德国 1.目标和假设 这是我们去年的CRCNS提案的重新提交，该提案获得了良好和非常好的分数， 审稿人。评论家们对建立在神经网络基础上的一般神经计算方法感到兴奋。 PI的补充技能。在这次修订中，我们解决了审稿人要求更清晰描述的问题 计划的实验和分析。重要的是，我们的新提案与临床 实践通过利用计算精神病学[1-4]和研究领域标准的思想， (RDoC例如，[5，6]），我们的目标是应用我们的神经计算方法来提高对 许多普遍性神经精神障碍所共有的核心社会缺陷。本提案的总体目标是 是建立全面的和临床相关的异常学习的神经计算模型 通过行为和功能性磁共振成像（fMRI）实验。 了解他人对于成功的社交互动至关重要[7]。强烈的社会互动 [2018 - 08 - 18]不同类型的社会功能障碍伴随着各种临床症状。 自闭症谱系障碍（ASD）的条件和构成核心症状[9-11]和人格 具有边界模式限定符（BPD）的疾病[12-16]。我们利用我们的神经计算方法 研究社会知识结构是如何形成的，反过来又是如何通过学习他人而形成的。 我们在计算中提出的知识表示和学习的基础机制 建模方法本身不是“社会”的，我们认为这是一个优势，他们可以应用于学习 在各种（非）社会领域。 在这里，我们专注于社会学习，以指定健康和健康之间的共性和差异。 个体和具有与ASD或BPD相关的显著社会缺陷的个体。虽然这两 临床组可能都采用过于僵化的社会知识结构，他们表现出不同的类型， ASD的特点是低估，即，关于心理的推论不足 他人的状态-可能是由于穷人或不具体的社会知识[17-19]。相比之下，BPD显示超过- 对他人的个性或意图进行心理化和过度负面的解释[12，13]。我们的新型 神经计算框架可以改善对社会功能障碍的理解，沿着维度 以及ASD和BPD的精神病分类标准。 我们的目标是测试和完善计算模型，形式化适当的策略，以获取 在学习过程中运用社会知识结构。我们的模型提供了规范的观点， 社会学习通过指定理想的代理如何在我们控制但生态有效的任务中学习。 因此，我们可以量化有多少人-特别是临床人群的特点是普遍存在的 社会赤字-偏离计算指定“最佳”基准策略。同时， 该项目旨在揭示社会知识表征背后的神经计算 在学习过程中灵活运用。该项目有三个具体目标：

PROJECT DESCRIPTION US-German Research Proposal for Collaboration in Computational Neuroscience: Efficient representations of social knowledge structures for learning from a computational, neural and psychiatric perspective (RepSocKnow) US-side PI: Prof. Gabriela Rosenblau, Ph.D., Assistant Professor, Department of Psychology, The George Washington University, 2115 G Street NW, Washington, DC 20052 German-side PI: Prof. Christoph W. Korn, Ph.D., Assistant Professor & PI Emmy-Noether Group, Section Social Neuroscience, Department of General Psychiatry, University of Heidelberg, Vossstraße 4, 69115 Heidelberg, Germany Consultant 1: Prof. Daniela Schiller, Ph.D., Associate Professor, Department of Psychiatry, Department of Neuroscience, and Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, 1470 Madison Ave, New York 10029, NY Consultant 2: Jan Gläscher, Ph.D., PI Bernstein Research Group, Institute for Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Martinistrasse 52, 20246 Hamburg, Germany 1. Aims and hypotheses This is a resubmission of our last year's CRCNS proposal, which received good and very good scores from reviewers. Reviewers were excited about the general neuro-computational approach that builds on the complementary skills of the PIs. In this revision, we address the reviewers' requests for clearer descriptions of the planed experiments and analyses. Importantly, our new proposal has direct relevance for clinical practice. By leveraging ideas from computational psychiatry [1–4] and from the Research Domain Criteria (RDoC; e.g., [5, 6]), we aim to apply our neuro-computational approach to improve the understanding of core social deficits shared by many pervasive neuro-psychiatric disorders. The general goal of this proposal is to establish comprehensive—and clinically relevant—neuro-computational models of aberrant learning in social contexts via behavioral and functional magnetic resonance imaging (fMRI) experiments. Learning about others is crucial for successful social interactions [7]. Social interactions strongly predict wellbeing [8]. Different types of impairments in social functioning accompany a variety of clinical conditions and constitute core symptoms of Autism Spectrum Disorders (ASD) [9–11] and Personality Disorders with a Borderline pattern qualifier (BPD) [12–16]. We harness our neuro-computational approach to investigate how social knowledge structures shape—and in turn are shaped by—learning about others. The mechanisms underlying knowledge representations and learning that we propose in our computational modeling approach are not “social” per se and we deem it a strength that they can be applied to learning across various (non-)social domains. Here, we focus on social learning to specify commonalities and differences between healthy individuals and individuals with marked social deficits associated with ASD or BPD. While these two clinical groups probably both employ overly rigid social knowledge structures, they exhibit different types of malfunctioning: ASD are characterized by under-mentalizing, i.e., insufficient inferences about mental states of others—possibly due to poor or unspecific social knowledge[17–19]. In contrast, BPD show over- mentalizing and overly negative interpretations of others' personality or intentions[12, 13]. Our novel neuro-computational framework can improve the understanding social malfunctioning along dimensional and categorical psychiatric criteria of ASD and BPD. We aim to test and refine computational models that formalize adequate strategies for acquiring and employing social knowledge structures during learning. Our models offer normative perspectives on social learning by specifying how ideal agents could learn in our controlled but ecologically valid tasks. Thereby, we can quantify how much humans—particularly clinical populations characterized by pervasive social deficits—deviate from computationally specified “optimal” benchmark strategies. In conjunction, this project aims to reveal the neural computations underlying the representation of social knowledge structures and their flexible deployment during learning. The project addresses three specific aims: