NMFを用いた大脳皮質MSTd野におけるオプティカルフローのパターン抽出
スポンサーリンク
概要
- 論文の詳細を見る
Cells in the area MT and MST are considered to analyze the visual movements of objects. A large number of cells in the area MT are selective to orientations of motion, while a large number of cells in the area MSTd, which is in the dorsal division of the medial superior temporal cortex, respond to rotation, expansion, contraction, and translation motions, thus analyzing optical flows. The cells in the area MSTd receive the input from cells in the area MT. This work aims at modeling how such neurons in MSTd emerge, where we choose non-negative matrix factorization (NMF) as our model. The cells in the area MSTd receive the input from cells in the area MT. NMF is a method to learn parts-based representations as a non-negative matrix by using non-negativity constraints. Since macaque's area IT is known to recognize an object by a combination or activations of cortical columns each selective to a part of an object, and since the firing rates of neurons are never negative, it is suggested that NMF explains how the brain might learn the parts of objects. In order to demonstrate the possibility that NMF also explains the mechanism of area MSTd, I have applied block matching method to natural movies for taking the motion-data (optical flow), and have applied NMF to the data to learn the bases. Then I examined whether the bases contained motion-patterns such as rotation, expansion, contraction, and translation. By this procedure, we explain the information processing in MSTd in terms of NMF.
- 2010-03-02
著者
関連論文
- BESOMモデルに基づく階層的時系列の学習と認識
- BESOMを用いたスパース符号化の一手法
- NMFを用いた大脳皮質MSTd野におけるオプティカルフローのパターン抽出
- プロセスとチャネルに基づく並列言語のための部分計算
- NMFによるcIPS野における立体的特徴認識のモデル化
- 初期視覚野におけるBorder-ownership信号の確率論的解釈
- 抽象カテゴリ学習