Using eye-tracking and computational modeling to understanding the dynamic allocation of attention during category learning

人们善于学习区分。它们自然而然地出现，没有任何特别的刻意努力，来区分狗和猫，愤怒和悲伤，以及桌子和椅子。但是，并不是所有的东西都很容易识别。例如，放射科医生需要多年的培训来检测X光胶片中的肿瘤。这项建议中的研究旨在提高我们对人类如何学习对事物进行分类的理解。我们的实验和模型的目的是了解选择性注意，即注意重要特征和忽略无关特征的能力，如何实现分类，并与我们的记忆和感知系统相互作用。这一研究领域的理论是正式的理论，即数学公式，它精确地指明了一个人可能认为某个东西属于什么类别，给定了这个物体的样子和这个人知道什么。实验被用来记录人们在学习类别时的反应是如何改善的，这些数据被用来定量检验形式理论的预测。这项研究的主要目标是提出一种准确的选择性注意的形式理论。通过研究眼动，人们可以用直接研究注意力的困难来换取更容易处理的问题，这些问题可以在实验室里处理。眼睛的运动是由大脑控制的，它们的位置是从经验中学习的；当一个物体或位置对我们有用的时候，我们的视觉注意力就会被吸引到那个位置，随着注意力的转移，我们的凝视也会随之转移。传统上，科学家们要么专注于眼睛的生物学，要么关注注意力的研究，要么关注学习的问题，但很少在一个研究项目中将这三个研究交叉。我正在做的是把数学方面的专业知识和学习心理学结合起来，通过研究眼球运动来检验习得的注意力。我正在使用先进的眼睛跟踪设备、数据分析和计算机模拟来实现这一点，这是由加拿大的高性能计算集群实现的。作为认知科学的基础性研究，目前的工作重点是从总体上提高我们对人类认知的理解。然后，应用科学家可以利用这些知识来设计基于计算机的自动分类系统，或者改进专家分类员的培训程序，例如需要解读行李扫描仪输出的机场安检人员。

People are good at learning to make distinctions. They naturally come, without any special conscious effort, to distinguish between dogs and cats, anger and sadness, and tables and chairs. But, not everything is easy to identify. Radiologists, for example, need years of training to detect tumors in x-ray films. The research in this proposal is aimed at improving our understanding of how humans learn to classify things. The purpose of our experiments and models is to understand how selective attention, the ability to pay attention to important features and ignore irrelevant ones, enables categorization and interacts with our memory and perceptual systems. Theories in this area of research are formal theories, mathematical formulas that precisely specify what category a person is likely to think something belongs to, given what the object looks like and what the person knows. Experiments are used to record how people's responses improve when learning categories, and these data are used to quantitatively test the predictions of formal theories. The primary goal of the proposed research is to produce an accurate formal theory of selective attention. By studying eye-movements one trades the difficulties in studying attention directly for more manageable problems that can be dealt with in the lab. Eye-movements are controlled by the brain, and their positioning is learned from experience; when an object or location is useful to look at, our visual attention is drawn to that location, and with attention goes our gaze. Scientists have traditionally focused on either the biology of the eyes, the study of attention or the problem of learning but rarely do the three intersect in a single research program. What I am doing is bringing together expertise in mathematics and the psychology of learning to examine learned attention by studying eye-movements. I am doing this using sophisticated eye-tracking equipment, data analysis and computer simulations made possible by Canada's high-performance computing clusters. As basic research in cognitive science, the present work is focused on improving our understanding of human cognition generally. This knowledge can then be used by applied scientists to design automatic computer-based classification systems or to improve training procedures for expert classifiers, such as airport security personnel who need to interpret the output of luggage scanners.