Optimal neural and behavioral markers for learning to learn during infancy

描述（由申请人提供）：人类婴儿面临着一个充满歧义的复杂世界。环境中不仅存在许多不同的信息特征和维度，而且这些线索通常与任何强化或反馈无关。在复杂和模糊的环境中学习有两种解决方案：（a）对选择用于处理的线索的先天约束（自下而上），或（B）评估线索的快速学习机制（自上而下）。学习的自上而下的信息选择机制可能更适合特定的任务需求，因此对学习更有用。考虑到婴儿在生命的头两年需要学习的东西，主要使用缓慢但精确（自上而下）的搜索方法是没有效率的。我的假设是，学习如何学习的发展进程需要以系统的方式使用自下而上的信息，同时创建自上而下的缓冲区以防止自下而上的分心。研究计划中的实验将测试这一假设，每个实验评估一个额外的学习水平。复杂的行为技术（即，台式和头戴式眼睛跟踪）和补充的现有技术神经成像方法（即，功能性近红外光谱[fNIRS]，通过头皮上的非侵入性光探针测量空间定位的神经激活），以及应用于婴儿眼球运动数据的数据挖掘工具，将研究婴儿如何从计算机显示器和自然环境中学习。 该研究计划有四个具体目标：1）建立一个新的，强大的学习方法，包括行为和神经测量，2）研究注意力部署如何最佳地改善学习，3）将学习范式应用于自然环境，4）进行微观分析并开发婴儿眼动的计算模型。培训部分的重点是学习在婴儿期研究中使用两种最先进的方法（头戴式眼动仪和fNIRS），并学习使用创新的数据挖掘工具来分析婴儿眼球运动的模式，将观看行为与认知能力联系起来。该培训计划对于申请人的职业目标至关重要，即确定学习学习的最佳策略，从而为有学习困难的人群提供培训方案。 这些发现将使更大的发展科学社区的研究人员受益，以及人工智能，感知学习，教育，动物学习，机器学习和进化心理学。这项工作将有助于对典型发展中选择性注意和学习动态的基本了解，这反过来又会为有学习困难的人群提供信息。

DESCRIPTION (provided by applicant): Human infants are confronted with a complex world that is filled with ambiguity. Not only are many different features and dimensions of information present in the environment, but these cues are often unrelated to any reinforcement or feedback. There are two solutions to learning in a complex and ambiguous environment: (a) innate constraints on the cues selected for processing (bottom-up), or (b) rapid learning-to-learn mechanisms that assess cues (top-down). Learned top-down mechanisms of information selection may be tuned more to specific task demands, and thus more useful for learning. Given how much infants have to learn over the first two years of life, it is not efficient to use mainly slow but precise (top-down) search methods. My hypothesis is that the developmental progression of learning how to learn requires using bottom-up information in a systematic way, while creating top-down buffers against bottom- up distraction. The experiments in the research plan will test this hypothesis, with each experiment evaluating an additional level of learning. Sophisticated behavioral techniques (i.e., both table- and head-mounted eye- tracking) and complementary state-of-the-art neuroimaging methods (i.e., functional near-infrared spectroscopy [fNIRS], measuring spatially-localized neural activation via non-invasive light probes on the scalp), as well as data mining tools applied to infant eye movement data, will examine how infants learn to learn from both computer displays and in naturalistic settings. There are four specific aims in this research program: 1) to establish a new, robust measure of learning with both behavioral and neural measures, 2) to investigate how attentional deployment can optimally improve learning, 3) to apply the learning paradigm to the natural environment, and 4) to conduct microanalyses on and to develop computational models of infant eye movements. The training component focuses on learning to use two state-of-the-art methods in infancy research (a head-mounted eye-tracker and fNIRS), and learning to use innovative data mining tools to analyze patterns of infant eye movements to link looking behavior to cognitive abilities. This training program is essential for the applicant's career goal of identifying the optimal strategies for learning to learn that will lead to training regimens for populations with learning difficulties. The findings will benefit researchers within the larger community of developmental science, as well as artificial intelligence, perceptual learning, education, animal learning, machine learning, and evolutionary psychology. This work will contribute to a foundational understanding of the dynamics of selective attention and learning in typical development, which in turn would inform populations with learning difficulties.