CompCog: Collaborative Research: Learning Visuospatial Reasoning Skills from Experiences

该项目使用人工智能(AI)的方法来更好地了解人们如何学习心理旋转等视觉空间推理技能，这是发展强大的数学和科学能力的关键因素。特别是，该项目提出了一种新的方法来量化不同视觉体验中包含的学习价值，使用可穿戴式摄像头和新的人工智能系统相结合，该系统可以从视频示例中学习视觉空间推理技能。该项目的结果不仅将推动人工智能的最先进水平，还将使研究人员能够衡量不同的现实世界视觉体验在帮助人们学习视觉空间推理技能方面的价值。例如，某些类型的物体游戏活动可能对帮助儿童学习某些视觉空间推理技能特别有价值。最终，这种新的测量方法可以用来识别儿童视觉空间推理困难的早期迹象，也可以帮助设计新的视觉空间训练干预措施，以促进儿童早期的数学和科学发展？S。本项目旨在回答的核心科学问题是：如何从第一人称视觉经验中学习视觉空间推理技能？这个问题将通过一种新的人工智能系统-心理成像引擎(MIME)--的计算实验来回答，该系统从视频例子中学习视觉空间推理技能，如心理旋转。培训数据将包括来自两个不同环境的第一人称可穿戴式摄像头视频，这两个环境对人类学习都很重要：婴儿操纵非结构化物体，以及为儿童设计的视觉空间培训干预。MIME人工智能系统的实验结果将推动人工智能和人类学习科学的发展，帮助解释如何从视觉体验中学习视觉空间推理技能，特别是不同类型的视觉体验如何影响一个人的素质？S以不同的方式学习结果。

This project uses methods from artificial intelligence (AI) to better understand how people learn visuospatial reasoning skills like mental rotation, which are a critical ingredient in the development of strong math and science abilities. In particular, this project proposes a new approach to quantify the learning value contained in different visual experiences, using wearable cameras combined with a new AI system that learns visuospatial reasoning skills from video examples. Results from this project will not only advance the state of the art in AI but also will enable researchers to measure how valuable different real-world visual experiences are in helping people to learn visuospatial reasoning skills. For example, certain types of object play activities might be particularly valuable for helping a child to learn certain visuospatial reasoning skills. Ultimately, this new measurement approach could be used to identify early signs of visuospatial reasoning difficulties in children and could also help in the design of new visuospatial training interventions to boost children?s early math and science development.The core scientific question that this project aims to answer is: How are visuospatial reasoning skills learned from first-person visual experiences? This question will be answered through computational experiments with a new AI system---the Mental Imagery Engine (MIME)---that learns visuospatial reasoning skills, like mental rotation, from video examples. Training data will include first-person, wearable-camera videos from two different settings that are both important for human learning: unstructured object manipulation by infants and visuospatial training interventions designed for children. Results from experiments with the MIME AI system will advance the state of the art in both AI and the science of human learning by helping to explain how visuospatial reasoning skills can be learned from visual experiences, and, in particular, how having different kinds of visual experiences can affect the quality of a person?s learning outcomes in different ways.