The planning of new compositional action sequences guided by interpretation of ambiguous sensory data in a novel drawing task

项目摘要/摘要 动物表现出一系列具有创造性、适应性和灵活性的行为。鸟类和灵长类动物重新调整用途 建造巢穴和工具的新材料；老鼠高效地构建导航捷径，人类进行泛化 掌握一种语言的知识才能有效地说另一种语言。这种动态创建新奇行为的能力 单一试验“依赖于成分计划，或通过重组而产生策略的心理过程 以前学习过的行为组件。至关重要的是，这取决于如何解释模棱两可的问题(和 关联的感觉数据)使用先验知识。缺乏研究的实验框架。 构图规划。为了解决这一新方法的迫切需要，这项提议将阐明神经 我在Freiwald实验室开发的一个新的绘图任务的机制，在这个任务中，猕猴绘制副本 前所未见的视觉形象。与先前对动作序列的研究不同的是，动作序列被记忆或 外部指导，在这项任务中，图纸必须是内部生成的，并取决于对 模糊的感觉数据。我将检验两个中心假设：(1)行为取决于成分 计划，基于行动和排序规则的先验知识，以及(2)灵活的额叶皮质活动 重新组合与动作和规则相对应的神经活动轨迹的“库”。第一个目标是测试 认为行为依赖于行为程序的组成计划的工作假设，或者 从习得的动作词汇表(例如，笔画表示“直线”或“圆弧”)和抽象词汇构建的程序 排序规则(即，更高顺序的过程，如“重复”、“连接”)。我将应用无人监督的模型拟合 触摸屏和视频行为数据以及正式比较替代模式的工具。第二个目标是 通过记录大规模神经来识别复杂图形背后的动态神经表示 额叶皮质的活动。我将验证这样的假设，即新奇的图画是以组合的形式呈现的 编码动作和排序规则的神经活动轨迹的库。第三个目标是使用微型 刺激以检验工作假设，即神经活动对计划的因果贡献是 在时间上和解剖上特定的，映射到行为的潜在结构的方式。我预测 在特定时空位置的神经轨迹的扰动将导致特定的、结构化的、 行为障碍。预期的结果是对神经活动如何支撑 通过对模糊感觉数据的解释来指导对新的复杂动作的计划。这一点意义重大 因为它能更好地理解大脑如何将结构化的先验知识运用到创造性思维中 推理和行为。这项研究具有创新性，因为它引入了一种新的行为范式 专注于内部生成的、目标导向的序列，并集成了来自认知的概念和工具 具有大规模电生理学的科学。这将推动我们机械地理解 基于程序的内部操作和符号知识的推理。

PROJECT SUMMARY/ABSTRACT Animals exhibit a remarkable array of creative, adaptive, and flexible behaviors. Birds and primates repurpose new materials to build nests and tools; rats efficiently construct navigational shortcuts, and humans generalize knowledge of one language to efficiently speak another. This ability to dynamically create novel behavior “in a single trial” depends on compositional planning, or mental processes that generate strategies by recombining previously learned behavioral components. Crucially, this depends on interpreting ambiguous problems (and associated sensory data) using prior knowledge. There is a dearth of experimental frameworks for studying compositional planning. To address this critical need for new approaches, this proposal will elucidate neural mechanisms in a novel drawing task that I have developed in the Freiwald lab, in which macaques draw copies of never-before-seen visual figures. In contrast to prior studies of action sequences that are memorized or externally guided, in this task drawings must be internally generated and depend on cognitive interpretation of ambiguous sensory data. I will test two central hypotheses: (1) that behavior depends on compositional planning, based on prior knowledge of actions and sequencing rules, and (2) that frontal cortical activity flexibly recombines a “library” of trajectories of neural activity corresponding to actions and rules. The first aim will test the working hypothesis that behavior depends on compositional planning of behavioral programs, or procedures built from a learned vocabulary of actions (i.e., like strokes for “line” or “arc”) and abstract sequencing rules (i.e., higher-order procedures, like “repeat”, “connect”). I will apply unsupervised model-fitting tools to touchscreen and video behavioral data and formally compare alternative models. The second aim is to identify the dynamic neural representations underlying complex drawings by recording large-scale neural activity in frontal cortex. I will test the working hypothesis that novel drawings are represented as combinations of a library of neural activity trajectories encoding actions and sequencing rules. The third aim is to use micro- stimulation to test the working hypothesis that the causal contribution of neural activity towards planning is temporally and anatomically specific in a manner that maps onto the latent structure of behavior. I predict that perturbation of neural trajectories at specific spatio-temporal locations will lead to specific, structured, behavioral perturbations. The expected outcome is an algorithmic account of how neural activity underlies the planning of novel complex actions guided by interpretation of ambiguous sensory data. This is significant because it leads to better understanding of how the brain deploys structured prior knowledge in creative reasoning and behavior. This research is innovative because it introduces a new behavioral paradigm focusing on internally-generated, goal-directed sequences, and integrates concepts and tools from cognitive science with large-scale electrophysiology. This will push the boundaries of our mechanistic understanding of reasoning that is based on internal manipulation of programs and symbolic knowledge.