Date of Award
Doctor of Philosophy (PhD)
Primates are capable of recognizing faces even in highly cluttered natural scenes. In order to understand how the primate brain achieves face recognition despite this clutter, it is crucial to study the representation of multiple faces in face selective cortex. However, contrary to the essence of natural scenes, most experiments on face recognition literatures use only few faces at a time on a homogeneous background to study neural response properties. It thus remains unclear how face selective neurons respond to multiple stimuli, some of which might be encompassed by their receptive fields (RFs), others not. How is the neural representation of a face affected by the concurrent presence of other stimuli? Two lines of evidence lead to opposite predictions: first, given the importance of MAX-like operations for achieving selectivity and invariance, as suggested by feedforward circuitry for object recognition, face representations may not be compromised in the presence of clutter. On the other hand, the psychophysical crowding effect - the reduced discriminability (but not detectability) of an object in clutter - suggests that an object representation may be impaired by additional stimuli. To address this question, we conducted electrophysiological recordings in the macaque temporal lobe, where bilateral face selective areas are tightly interconnected to form a hierarchical face processing stream. Assisted by functional MRI, these face patches could be targeted for single-cell recordings. For each neuron, the most preferred face stimulus was determined, then presented at the center of the neuron's RF. In addition, multiple stimuli (preferred or non-preferred) were presented in different numbers (0,1,2,4 or 8), from different categories (face or non-face object), or at different proximity (adjacent to or separated from the center stimulus). We found the majority of neurons reduced mean ring rates more (1) with increasing numbers of distractors, (2) with face distractors rather than with non-face object distractors, (3) at closer distractor proximity, and, additionally, (4) the response to multiple preferred faces depends on RF size. Although these findings in single neurons could indicate reduced discriminability, we found that each stimulus condition was well separated and decodable in a high-dimensional space spanned by the neural population. We showed that this was because neuronal population was quite heterogeneous, yet changing response systematically as stimulus parameter changed. Few neurons showed MAX-like behavior. These findings were explained by divisive normalization model, highlighting the importance of the modular structure of the primate temporal lobe. Taken together, these data and modeling results indicate that neurons in the face patches acquire stimulus discriminability by virtue of the modularity of cortical organization, heterogeneity within the population, and systematicity of the neural response.
Ebihara, Akinori F., "Normalization Among Heterogeneous Population Confers Stimulus Discriminability on the Macaque Face Patch Neurons" (2015). Student Theses and Dissertations. 278.