Learning the parts of objects by auto-association

Authors:
Xijin Ge;Shuichi Iwata
Affiliations:
Research into Artifacts, Center for Engineering (RACE), The University of Tokyo, Komaba 4-6-1, Meguro-ku, Tokyo 153-8904, Japan;Research into Artifacts, Center for Engineering (RACE), The University of Tokyo, Komaba 4-6-1, Meguro-ku, Tokyo 153-8904, Japan
Venue:
Neural Networks
Year:
2002

Citing 6
Cited 2

Separating similar complex Chinese characters by Walsh transform

Pattern Recognition
Neural networks and principal component analysis: learning from examples without local minima

Neural Networks
The Helmholtz machine

Neural Computation
Autoassociator-based models for speaker verification

Pattern Recognition Letters
A simple algorithm that discovers efficient perceptual codes

Computational and psychophysical mechanisms of visual coding
Nonlinear Autoassociation Is Not Equivalent to PCA

Neural Computation

Learning Image Components for Object Recognition

The Journal of Machine Learning Research
Two-dimensional non-negative matrix factorization for face representation and recognition

AMFG'05 Proceedings of the Second international conference on Analysis and Modelling of Faces and Gestures

Quantified Score

Hi-index	0.00

Visualization

Abstract

Recognition-by-components is one of the possible strategies proposed for object recognition by the brain, but little is known about the low-level mechanism by which the parts of objects can be learned without a priori knowledge. Recent work by Lee and Seung (Nature 401 (1999) 788) shows the importance of non-negativity constraints in the building of such models. Here we propose a simple feedforward neural network that is able to learn the parts of objects by the auto-association of sensory stimuli. The network is trained to reproduce each input with only excitatory interactions. When applied to a database of facial images, the network extracts localized features that resemble intuitive notion of the parts of faces. This kind of localized, parts-based internal representation is very different from the holistic representation created by the unconstrained network, which emulates principal component analysis. Furthermore, the simple model has some ability to minimize the number of active hidden units for certain tasks and is robust when a mixture of different stimuli is presented.