一般口演（Oral）

Oral Vision-Ⅲ 一般口演視覚-Ⅲ

7月27日（土）17:30～17:45　第10会場（万代島ビル　6F　会議室） 3O-10e2-1 V4野における図地の符号化---単一細胞の受容野構造と集団符号化 Ko Sakai（酒井宏）¹，Kouji Kimura（木村耕治）¹，Yukako Yamane（山根ゆか子）^2,3，Hiroshi Tamura（田村弘）^2,4 ¹筑波大　システム情報 ²大阪大院生命機能 ³日本学術振興会 ⁴脳情報通信融合研究センター Scene segmentation, or Figure-ground (FG) segregation, is a fundamental step toward object recognition and scene understanding. We investigated the responses of V4 neurons to signal figures and ground, with a focus on local image information contained in natural images without explicit global context. We recorded spiking activities of a population of macaque V4 neurons while a variety of natural image patches and their variants were presented, and intended to evaluate how V4 neurons represent FG information. Around one third of the visually responsive neurons showed response modulation depending on the positional relation between the classical receptive field (CRF) of the neuron and the figural region of the stimulus. First, we estimated the spatial structure of the RFs using spike triggered stimulus average (STSA). With our aim of detecting the spatial organization of the RFs in response to FG configuration, we linked the FG to brightness. We grouped the stimulus patches based on their luminance contrast of the figure region with respect to the ground region, and averaged separately for the figure-bright and figure-dark groups. Then, we generated STSAs for each group and took the difference between the two. We compensated nonuniformity of luminance in the natural images by subtracting the simple ensemble average of the stimuli from the STSA. Estimated RFs exhibited antagonistic structures; facilitative and suppressive regions on the preferred and non-preferred figure sides, respectively. This result indicates the detection of local FG configuration by individual neurons. Next, we estimated the consistency of individual neurons in FG discrimination across a variety of natural patches. The consistency was approximately 55%, indicating a low capability of individual neurons in judging FG across a variety of natural patches. This result might not be surprising given a wide variety of natural images. We then hypothesized that an integration of responses across a small population of FG-modulated neurons would be capable of correctly judging FG across a variety of images. To test this hypothesis, we used the support vector machine to integrate the responses of multiple neurons, and computed the consistency. The integration of the activities of a few tens of neurons yielded the discrimination consistency far greater than that of single neurons (up to 72%), suggesting a distributed representation of FG information in V4.		7月27日（土）17:45～18:00　第10会場（万代島ビル　6F　会議室） 3O-10e2-2 下側頭葉皮質TE野とTEO野における三次元物体表現の違い Jun-Ya Okamura（岡村純也），Koki Uemura（上村洸樹），Lulin Dai（戴璐璘），Gang Wang（王鋼）鹿児島大院理工情報生体 Previous behavioral studies showed monkeys acquired across-view object recognition capability in a certain range of viewing angle after the discrimination experience between similar objects at the same viewing angle without association of different views. Following electrophysiological study found neurons in area TE, which is the final visual area along the ventral visual pathway, show response tolerance in a viewing angle range of 30-60° to the objects experienced in discrimination at the same viewing angle. In the present study to understand the underlying neural mechanism of such stimulus selectivity of TE cells, responses of neurons in area TEO, which supply main inputs to area TE, to the same experienced object images were recorded. Four similar objects were created by deforming a prototype object. Four views were created by rotating the objects in depth from 0° to 90° in 30° intervals. A total of 16 object images formed an object set. To the same object sets, we found single cells in area TEO did not show any viewing angle tolerance to the objects experienced in within-view discrimination task, while TE cells of the same animal did. Population activity to a single stimulus was evaluated by pooling responses of single cells to the same stimulus into a vector. For each pair of object images, we computed the correlation coefficient between the vectors. Neural distances for a pair of images were defined by subtracting the correlation coefficient from 1. Neural distances for views of the same objects at a viewing angle difference of 30° were significantly smaller than those for views across different objects in an object set in both areas TE and TEO after experiencing the object images in the within-view discrimination task. However, the differences of the neural distances for views of the same objects and different objects were significant at viewing angle differences of 60 and 90° only in area TE, and not significant in area TEO. The results suggest that, in the level of cell population, the representation of the three dimensional objects in area TE is more view-invariant comparing to those in area TEO.

7月27日（土）18:00～18:15　第10会場（万代島ビル　6F　会議室） 3O-10e2-3 顔と顔以外の分類および顔による個体の識別に必要なサル下側頭葉視覚連合野の神経集団の違い Yuko Nirei（楡井優子）^1,2，Go Uchida（内田豪）¹，Manabu Tanifuji（谷藤学）^1,2,3 ¹理研CBS 脳統合機能 ²早稲田大院先進理工生医 ³東京大院新領域創成科学複雑理工 Visual object recognition generally has to solve two problems: classification (assigning objects into proper object categories) and identification (specifying identity of objects). The inferior temporal (IT) cortex is a visual area crucial to visual perception and recognition of objects. A class of neurons in IT cortex (face neurons) preferentially responds to faces and is supposed to detect visual features of face images. In the present study, we addressed a question whether visual features detected by face neurons serve enough information for both face classification and identification of identities based on faces. We identified neurons representing visual features required for discrimination between faces and non-face objects (face classification), or between faces of two identities (face identification). We penetrated electrodes into IT cortex and recorded neural responses to 1550 objects (819 faces and 731 non-face objects) from 413 sites. The face images include faces of 6 identities and each identity has 20 images with different views. We defined face sites as the sites where responses of neurons have significant positive correlation with binary responses: one to faces and zero to non-face objects, and non-face sites as the rest of the sites. We found 331 face sites and 82 non-face sites. We conducted face classification, and face identification (in total 15 pairs) by L1 regularized logistic regression. Logistic regression is a two-class-classification algorism in machine learning. In this study, the classifier is represented by a weighted sum of responses at sites. In L1 regularized logistic regression, the weights of sites not necessary required in the discrimination result in zero. Neurons at sites with positive or negative weights represent visual features included in class 1 or class 2, respectively. In the face classification, faces are labeled as class1 and non-face objects as class2. We found sites with positive weights from face sites and those with negative weights from non-face sites. In the face identification, we labeled faces of one identity as class 1 and the other as class2. We found both sites with positive or negative weights from both face sites and non-face sites. Our results suggest that visual features included in faces are represented by neurons at face sites; on the other hand, visual features required for identification of identities by faces are represented by neurons not only at face sites, but also at non-face sites.		7月27日（土）18:15～18:30　第10会場（万代島ビル　6F　会議室） 3O-10e2-4 サルに適用可能な素材弁別課題による素材評価と触感の評価の関連性 Minami Ito（伊藤南）¹，Tensei Iwata（岩田天晴）²，Kanoko Katsube（勝部夏乃子）²，Yuka Morisue（森末有香）² ¹東京医歯大医歯総合生体機能システム学 ²東京医歯大医学部保健衛生学科 Materials of surface provides a powerful clues to object recognition by perceiving both visual texture information and secondary haptic information like roughness or hardness. However, the haptic perception might be dependent on animal's sensation and experiences, and the property is not well understood in model animals used for physiological studies. Though we have trained monkeys to perform the material discrimination task, it is hard to ask about the haptic sensation in these animals. Here, we used the same discrimination task, which might be easy for human subjects but very hard for non-human subjects. Then, we examined the relationships between the scores of the material discrimination and the haptic discrimination in human subjects. Thirteen human subjects (6 male and 7 females, age 21-57) conducted two to four daily sessions of the task. In the haptic discrimination task, subjects reported 4 haptic scores (rough/smooth, hard/soft, hot/cold or dry/wet) in 5 steps for each material. In the material discrimination task, subjects reported 5 scores indicating resemblance with 5 reference materials (metal, wood, carpet, gel-sheet, and fur). The task was conducted with or without visual inspection. Order of sample presentation was randomized and each sample was presented once in a daily session. On the other hand, three Japanese monkeys (macaca fuscata, female, 7.0, 6.8 & 6.9kg) has been trained with the material discrimination paradigm over 24, 18 & 6 months. After pressing a target material, they had to touch the same one among 5 reference materials. After the performance has improved above 90%, new object was presented once in a daily session. They have to choose one among 5 reference materials as a response of their categorization. We compared these scores in two dimensional spaces after the principle component analysis following multi-dimensional scaling procedures. We found that (1) scores of the material and haptic discrimination tend to form compact cluster for each material categories, (2) linear regression sufficiently predict haptic scores from the material scores (p<0.05), (3) material categorization of primate-subjects was largely consistent with human subjects. In conclusion, our material discrimination paradigm has a potential to evaluate material perception within and between the material categories, expecting direct comparison between human and non-human primate subjects.