Rapid training of higher-order neural networks for invariant pattern recognition

摘要: The authors demonstrate a second-order neural network that has learned to distinguish between two objects, regardless of their size or translational position, after being trained on only one view each object. Using an image 16*16 pixels, the training took less than 1 min run time Sun 3 workstation. A recognition accuracy 100% was achieved by resulting for several test-object pairs, including standard T-C problem, any position and over scale factor five. takes advantage known relationships input pixels build invariance into architecture. use third-order achieve simultaneous rotation, scale, is described. Because high level rapid, efficient training, initial results show higher order networks be vastly superior multilevel first-order backpropagation applications where invariant pattern required. >