We present a new classification architecture based on autoassociative
neural networks that are used to learn discriminant models of each cla
ss. The proposed architecture has several interesting properties with
respect to other model-based classifiers like nearest-neighbors or rad
ial basis functions: it has a low computational complexity and uses a
compact distributed representation of the models. The classifier is al
so well suited for the incorporation of a priori knowledge by means of
a problem-specific distance measure. In particular, we will show that
tangent distance (Simard, Le Cun, & Denker, 1993) can be used to achi
eve transformation invariance during learning and recognition. We demo
nstrate the application of this classifier to optical character recogn
ition, where it has achieved state-of-the-art results on several refer
ence databases. Relations to other models, in particular those based o
n principal component analysis, are also discussed.