Visual Adaptation and Neuronal Selectivity

现在可以设计每秒计算数百万次、执行有用任务和识别对象的计算机。然而，尽管有这些进步，计算机在灵活性方面还不能模拟大脑的功能。大脑的一个显著特点是，它以高度依赖情境和灵活的方式来解释刺激和组织行动。对于视觉刺激，大脑能够学习大量刺激类别的结构和意义。对于某些类别，例如人脸，它的性能是非常显著的：我们可以很容易地根据人脸成分及其几何形状的非常细微的差异来区分和识别数千个不同的人。考虑到我们每次看到一张给定的面孔，它在我们视网膜上的图像与上次不同，这一点更加令人印象深刻。两个不同的人从相同的距离和相同的照明条件下看到的视网膜图像可能非常相似，至少在粗略的水平上，与同一个人在不同条件下两次看到的图像非常相似。然而，我们能够流畅和毫不费力地识别人、物体、地标和场景，基于一眼。这种能力是如何产生的？ 这个问题的一个答案与复杂的视觉刺激在大脑中的编码方式有关。这个话题一直是我们研究的中心焦点。在过去的一年里，我们发表了关于视觉皮质V4和TE区域中刺激的神经表征的论文。我们之前已经证明，单个人脸是基于它们相对于平均人脸的区别性进行系统编码的，即所谓的基于范数的编码。换句话说，大脑根据其结构与刻薄的原型面孔的不同，对给定的面孔进行编码。我们首先通过进行涉及视觉适应的人类行为实验，为这种编码方式提供了证据。在这些实验中，一张脸呈现几秒钟会改变随后呈现的脸被感知的方式。这些误解与基于规范的编码的预期非常吻合。最近对非人类灵长类动物的神经生理记录加强了这一点，该记录显示，颞下皮质中的神经元根据一张脸与多张脸的平均值的相对差异来调整其放电频率。这两条研究路线都指向这样一个结论，即大脑系统地编码面孔身份，并相对于典型的平均水平。 面孔感知的第二个重要特征是学习和记住新面孔的能力，这一过程无疑涉及大脑的变化。与一些技能(如语言习得)不同，我们学习新面孔的能力在成年后仍然很强。这意味着，在某种意义上，我们认知的神经机制仍然是可塑性的。体验是如何改变神经反应的？在过去的一年里，我们已经开始通过监测单个神经元在几天和几周的时间内的调谐功能来解决这个问题。虽然从单个细胞进行记录是一个常规过程，但对它们进行长时间的监测会带来巨大的挑战。我们已经克服了这些挑战，在外部合作者的帮助下开发了一种新型的无惯性微电极束阵列，它通过随着大脑的微小运动而保持与单个神经元的紧密接近。这种方法的优点是，在记录多天时间内分离神经元的反应时，可以评估视觉学习对神经选择性的影响。在实验室里，随着神经反应的监测，非人类灵长类动物目前正在接受训练，学习新类别的刺激，包括新的人类和猿人面孔。这项研究的结果将阐明我们如何能够学习刺激，因为视觉神经元的选择性发生了变化。

It is now possible to design computers that compute millions of calculations per second, perform useful tasks, and recognize objects. However, despite these advances, computers fall short of emulating brain function in the domain of flexibility. One of the remarkable aspects of the brain is that it interprets stimuli and organizes its actions in a highly situation-dependent and flexible manner. With regard to visual stimuli, the brain is able to learn the structure and significance of a large number of stimulus categories. For some categories, such as faces, its performance is utterly remarkable: we readily can discriminate between and recognize thousands of different individuals based on very subtle differences in the face components and their geometrical configuration. This is all the more impressive given that each time we see a given face, its image on our retina is different from the last time. Two different individuals seen from the same distance and same lighting conditions may cast very similar retinal images, at least at a coarse level, than the same individual seen twice under different conditions. Nonetheless, we are able to fluidly and effortlessly recognize people, objects, landmarks, and scenes based on a single glance. How does this ability come about? One answer to this question relates to the manner in which complex visual stimuli are encoded in the brain. This topic has been a central focus of our research. In the past year, we have published papers related to the neural representation of stimuli in object-encoding regions V4 and TE of the visual cortex. We have previously shown that individual faces are systematically encoded based on their distinctiveness relative to an average face, so called norm-based encoding. In other words, the brain encodes to a given face according to how it differs in its structure from a mean, prototypical face. We first provided evidence for this means of encoding by conducting human behavioral experiments involving visual adaptation. In those experiments the presentation of one face for a few seconds altered the way a subsequently presented face was perceived. The misperceptions closely matched the expectations of norm-based encoding. This was strengthened by more recent neurophysiological recordings in nonhuman primates showing the neurons in the inferotemporal cortex adjust their firing rate based on the relative difference of a face from the average of many faces. Both lines of research point to the conclusion that the brain encodes face identity systematically, and relative to a prototypical average. A second important feature of face perception is the ability to learn and remember new faces, a process that undoubtedly involves changes in the brain. Unlike some skills (e.g. language acquisition), our capacity to learn new faces remains strong into adulthood. This implies that the neural machinery underlying our recognition remains, in a sense, plastic. How does experience modify neural responses? During the past year we have begun to approach this problem by monitoring the tuning functions of individual neurons over periods of days and weeks. While recording from single cells is a routine process, monitoring them for extended periods of time poses enormous challenges. We have overcome these challenges by developing, with the help of an outside collaborator, a novel inertialess microelectrode bundle array, which maintains close proximity to individual neurons by moving with the small movements of the brain. The advantage of this approach is that in recording the responses of isolated neurons over a period of many days the effects of visual learning on neural selectivity can be assessed. In the laboratory, nonhuman primates are presently being trained to learn new categories of stimuli, including novel human and simian faces, as neural responses are monitored. The results from this study will shed light on how we are able to learn stimuli because of changes in the selectivity of visual neurons.