Analysis of Representations of 3-Dimensional Objects in the Cell Populations in the Temporal Association Area Using Machine Learning
Analysis of Representations of 3-Dimensional Objects in the Cell Populations in the Temporal Association Area Using Machine Learning
复制标题
使用机器学习分析时间关联区域细胞群中 3 维物体的表示
DOI：
10.1007/978-3-030-66169-4_23
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Wang Gang
中科院分区：
文献类型：
--
作者：
Okamura Jun-ya;Yamamoto Yusuke;Dai Lulin;Uto Yoshihiro;Yamada Yousuke;Wang Gang
三维物体可以被识别，而不管视角如何。有研究表明，在一定的视角范围内，颞叶联合区的细胞对在相同视角下辨别所经历的物体有耐受性反应。在本研究中，机器学习被应用到时间关联区域的群体响应向量。训练分类器以创建在相同视角下将对象与相似对象分离的超平面，然后通过对不同视角下的对象图像的群体响应向量进行测试。从真阳性、假阳性和d'值评估的辨别性能在整个100-600 ms时间窗口中在相同视角下经历辨别的对象中高于没有辨别经历的对象。在同一视角下，在100-220 ms时间窗的反应早期，d'值较低，而在220-600 ms时间窗的反应晚期，d'值较高。通过学习对物体的不同观点的联想而经历的物体的d'值在反应的早期和晚期都很高。结果表明，计算模型的对象表示创建通过使用在时间关联区域的细胞反应，和不同的时间窗口中的视角公差的表示的差异。
Three-dimensional objects can be recognized regardless of the viewing angles. It was reported that the cells in the temporal association area responded tolerantly in a certain viewing angle range to the objects experienced in discrimination at the same viewing angles. In the present study, machine learning was applied to the population response vectors in the temporal association area. A classifier was trained to create a hyperplane that separated an object from similar object at the same viewing angles and then tested by the population response vectors to the object images at a different viewing angle. The discrimination performance evaluated from true positives, false positives, and d’ values was higher in the objects experienced in discrimination at the same viewing angles than those without discrimination experience in the whole 100–600 ms time window. For the objects experienced in discrimination at the same viewing angles, the d’ values were low in the early phase of the responses of 100–220 ms time window, but became higher in the late phase of 220–600 ms time window. The d’ values for the objects experienced by learning association of different views of the objects were high in the early and late phases of the responses. The results demonstrate computational models for object representation created by using cellular responses in the temporal association area, and differences of the representation of the viewing angle tolerance in different time windows.
识别从不同角度看到的物体不需要旋转经验
DOI：
--
发表时间：
2005
期刊：
Nature Neuroscience 8
影响因子：
--
作者：
G.Wang;S.Obama;W.Yamashita;T.Sugihara;K.Tanaka
通讯作者：
K.Tanaka