PERCEPTUAL LEARNING: HUMAN VS. OPTIMAL BAYESIAN

描述（由申请人提供）：人类在大范围视觉任务中的表现随着练习而提高。观察者的准确性提高的一个过程是由于他们使用任务相关（信号）信息的能力的提高。然而，调节这一学习过程的人类神经算法的动态是未知的。我们假设一个新的实验范式可以通过将经验测量的人类感知学习性能与最优贝叶斯学习器和其他次优学习模型进行比较，帮助阐明人类学习算法，并确定学习效率低下的人为来源。为了实现这一目标，我们提出了四个具体目标：(1)开发一个实验范式，使研究者能够将人类感知学习与最优贝叶斯学习者的学习以及其他次优学习模型进行比较，以完成涉及学习不同视觉属性的五种特定任务；(2)开发一系列诊断测试，以比较人类学习和模型学习；(3)从心理物理角度测量五种视觉任务的人类视觉表现，并使用一系列诊断测试将其与模型表现进行比较；(4)利用开发的实验和理论框架来评估人类使用不同模式的学习反馈的效率。这项工作将提高我们对调节成人感知学习动态的人类神经算法的理解，并确定人类学习效率低下的不同来源。最后，提出的实验方案和理论框架将提供一个新颖、强大和灵活的框架，其他研究人员可以使用它来评估正常成人和婴儿在各种任务和感觉模式下的感知学习，评估有视觉障碍和/或学习障碍的人类的学习，并可能与细胞记录和/或脑成像结合使用。

DESCRIPTION (provided by applicant): Human performance in a large range of visual tasks improves with practice. One process by which observers' accuracy increases is due to an improvement in their ability to use task-relevant (signal) information. However, the dynamics of the human neural algorithm mediating this learning process is mostly unknown. We hypothesize that a new experimental paradigm can help elucidate the human learning algorithm as well as identify the human sources of learning inefficiency by allowing comparisons of empirically measured human perceptual learning performance to that of an optimal Bayesian learner and other suboptimal learning models. To achieve this goal we propose four Specific Aims: (1) to develop an experimental paradigm that allows the investigator to compare human perceptual learning to the learning of an optimal Bayesian learner as well as other suboptimal learning models for five specific tasks involving learning about different visual attributes; (2) to develop a battery of diagnostic tests to compare human and model learning; (3) to measure human visual performance psychophysically for the five proposed visual tasks and compare them to model performance using the battery of diagnostic tests; and (4) to use the developed experimental and theoretical framework to evaluate the efficiency of the human use of different modes of feedback on learning. The proposed work will improve our understanding of the human neural algorithms mediating the dynamics of adult perceptual learning and identify different sources of inefficiency in human learning. Finally, the experimental protocols and theoretical framework proposed will provide a novel, powerful and flexible framework that other researchers can use to evaluate normal adult and infant perceptual learning in a variety of tasks and sensory modalities, to assess learning in humans with visual disorders and/or learning disabilities, and could potentially be used in conjunction with cell recording and/or brain imaging.