Learning to Recognize Faces Despite Within-Person Variability in Appearance: A Developmental Approach

人脸识别是一项挑战：同一个人的两张照片可能看起来非常不同，而两个人的照片可能看起来非常相似。尽管如此，尽管人的外表存在差异，成年人仍然可以识别成千上万张面孔——我们依靠这种能力在繁忙的火车站找到朋友，跟上电影情节，和邻居打招呼。这种识别熟悉面孔的技能造成了对所有面孔都具有专业知识的错觉，证明在涉及不熟悉面孔的情况下使用带照片的身份证件是合理的（例如，在获得医疗保健或购买年龄限制商品时；边境安全）。当被问及不熟悉的面孔的照片是属于同一个人还是两个不同的人时，成年人会犯很多错误；灯光或表情的简单变化可能会被误解为身份的变化。大量的研究正在研究成年人对陌生面孔和熟悉面孔的心理表征，以及新面孔变得熟悉的过程。机器学习的重大进展为人脸识别的理论模型提供了信息。发展研究对于丰富我们对成人专业知识的理解、改进机器学习以及识别儿童犯错的条件至关重要。我发现，在童年时期，对不熟悉的面孔进行身份匹配的能力会有一个漫长的发展过程，就像学习新面孔的能力一样。到6岁，但不是更小的时候，孩子们可以很容易地认出非常熟悉的面孔（他们的父母/老师）。根据最先进的机器学习算法，我的提案包括三个项目，以研究儿童时期面部学习和识别的长期发展。项目A使用头戴式摄像机来发现儿童在与成人和同龄人互动时对面部（相对于物体）的注意力与年龄相关的变化。我将研究经验是否会影响儿童在陌生面孔中匹配身份和识别高度熟悉面孔的能力。项目B关注与年龄相关的对关键特征（如嘴唇厚度、眼睛颜色）敏感性的变化——这些特征决定了成年人对陌生和熟悉面孔的身份感知。我将研究儿童对关键特征的敏感度是否低于成人，以及儿童和成人面部的关键特征是否不同。项目C将研究情境线索对儿童面部学习和识别的影响。我将发现儿童是否比成人更能从传记信息、在群体中学习面孔以及在学习环境中看到熟悉的面孔中获益。研究结果将提供关于儿童时期面部识别如何精细调整的新见解，为理论模型提供信息，并导致更好的人工智能。

Recognizing faces is challenging: Two pictures of the same person can look very different and pictures of two people can look very similar. Nonetheless, adults can recognize thousands of faces despite within-person variability in appearance-an ability we rely on to find a friend in a busy train station, follow movie plots, and greet neighbours. This skill for recognizing familiar faces creates the illusion of expertise for all faces, justifying the use of photo ID in settings that involve unfamiliar faces (e.g., when accessing health care or buying age-restricted goods; border security). Adults make many errors when asked whether photos of unfamiliar faces belong to the same person or two different people; a simple change in lighting or expression can be misperceived as a change in identity. A spate of research is examining adults' mental representation of unfamiliar vs. familiar faces and the processes by which a newly encountered face becomes familiar. Significant advances in machine learning have informed theoretical models of face recognition. Developmental research is essential to enrich our understanding of adults' expertise, improve machine learning, and identify conditions under which children make errors. I discovered that matching identity for unfamiliar faces has a prolonged development during childhood, as does the ability to learn a newly encountered face. By 6 years of age, but not younger, children recognize highly familiar faces (their parent/teacher) with ease. Informed by state-of-the-art machine learning algorithms, my proposal comprises three projects to examine the prolonged development of face learning and recognition during childhood. Project A uses head-mounted cameras to discover age-related changes in children's attention to faces (vs. objects) when interacting with adults and peers. I will examine whether experience impacts children's ability to match identity in unfamiliar faces and recognize highly familiar faces. Project B focuses on age-related changes in sensitivity to critical features (e.g., lip thickness, eye colour)-features that govern adults' perception of identity in both unfamiliar and familiar faces. I will examine whether children are less sensitive to critical features than adults and whether critical features differ for child vs. adult faces. Project C will examine the influence of contextual cues on children's face learning and recognition. I will discover whether children benefit more than adults from biographical information, learning faces in groups, and seeing familiar faces in the context in which they were learned. The results will provide novel insights about how face recognition becomes finely tuned during childhood, informing theoretical models and leading to better artificial intelligence.