A unified theory for perception of physical and social dynamics

人类自发地赋予我们所看到的世界以物理和社会意义：我们不仅“看到”无生命物体之间的物理相互作用，而且还看到相互作用的人类的有意行为。该项目将研究人类如何从物体和人的运动中感知和理解物理和社会动态。更大的目标是解释感知和推理如何共同作用，以指导动态情况下的行为。该项目的发现将有助于为下一代智能系统提供在新情况下操作所需的视觉推理能力。这样的系统将需要执行复杂的任务，这对人类来说是具有挑战性和耗时的。该项目将整合行为实验与计算建模，以开发和测试一个统一的理论。该理论详细说明了如何从图像中直接构建力表征，以及如何使用所发现的力结构来推断动态刺激中涉及的因果关系。具体来说，贝叶斯模型将使用两个基本计算组件来开发：（1）生成模型，用于学习控制物理和社会场景中无生命物体和人类代理运动的潜在力结构的表示，以及（2）推理模型，用于识别因果关系和隐藏状态基于力结构的时空配置的意图。计算模型将用于合成动态刺激，实例化大范围的物理和社会情况。然后，这些刺激将用于大规模分析人类行为。一系列心理物理实验和模型比较将测试该理论的关键计算组件。该项目还将为那些对认知科学和人工智能之间的跨学科研究感兴趣的学生提供极好的培训机会。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

Humans spontaneously endow the world we see with physical and social meaning: we “see” not only physical interactions among inanimate objects, but also the intentional behaviors of interacting humans. This project will investigate how humans perceive and understand physical and social dynamics from the movements of objects and people. The larger goal is to explain how perception and reasoning operate together to guide behavior in dynamic situations. Findings from this project will help equip the next generation of intelligent systems with the visual reasoning ability required to operate in novel situations. Such systems will need to perform complex tasks that are challenging and time-consuming for humans.The project will integrate behavioral experiments with computational modeling to develop and test a unified theory. The theory specifies how force representations can be constructed directly from images and how the discovered force structures can be used to infer causal relations involved in dynamic stimuli. Specifically, Bayesian models will be developed with two basic computational components: (1) generative models to learn representations of latent force structures governing movements of inanimate objects and human agents in physical and social scenes, and (2) reasoning models to identify causal relations and hidden states of intentions based on spatial-temporal configurations of force structures. Computational models will be used to synthesize dynamic stimuli that instantiate a large range of physical and social situations. These stimuli will then be used to analyze human behaviors at a large scale. A series of psychophysical experiments and model comparisons will test the key computational components of the theory. The project will also provide excellent training opportunities for students who are interested in interdisciplinary research at the interface between cognitive science and artificial intelligence.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.