CiNet Monthly Seminar (Special) : 福島邦彦, Deep CNN Neocognitron for Visual Pattern Recognition

CiNet Monthly Seminar

2016年11月22日(金)
16:00 ~ 17:30
会場 : CiNet 1F 大会議室

“Deep CNN Neocognitron for Visual Pattern Recognition”

福島邦彦
http://personalpage.flsi.or.jp/fukushima/

担当 :  Ferdinand Peper (PI)

Abstract:
Recently, deep convolutional neural networks (deep CNN) have become very popular in the field of visual pattern recognition. The neocognitron, which was first proposed by Fukushima (1979), is a network classified to this category. Its architecture was suggested by neurophysiological findings on the visual systems of mammals. It is a hierarchical multi-layered network. It acquires the ability to recognize visual patterns robustly through learning.

Although the neocognitron has a long history, improvements of the network are still continuing. This talk discusses the recent neocognitron focusing on differences from the conventional deep CNN. For training intermediate layers of the neocognitron, the learning rule called AiS (Add-if-Silent) is used. Under the AiS rule, a new cell is generated and added to the network if all postsynaptic cells are silent in spite of non-silent presynaptic cells. The generated cell learns the activity of the presynaptic cells in one-shot. Once a cell is generated, its input connections do not change any more. Thus the training process is very simple and does not require time-consuming repetitive calculation. In the deepest layer, a method called Interpolating-Vector is used for classifying input patterns based on the features extracted by the intermediate layers.

Some other functions of the visual system can also be realized by networks extended from the neocognitron, for example, mechanism of selective attention, recognition and completion of partly occluded patterns, restoring occluded contours, and analysis of optic flow.

About CiNet’s Monthly Seminar:
In CiNet Monthly Seminar, distinguished external (national and international) speakers will give a lecture about their latest work. The lecture will be held once a month from 16:00 at the CiNet main lecture theatre, followed by an informal discussion in the coffee room. List of speakers will be announced in advance. Everyone is welcome to attend, and welcome to join the after seminar discussion.