Part of Advances in Neural Information Processing Systems 24 (NIPS 2011)
Salah Rifai, Yann N. Dauphin, Pascal Vincent, Yoshua Bengio, Xavier Muller
We combine three important ideas present in previous work for building classi- ﬁers: the semi-supervised hypothesis (the input distribution contains information about the classiﬁer), the unsupervised manifold hypothesis (data density concen- trates near low-dimensional manifolds), and the manifold hypothesis for classiﬁ- cation (different classes correspond to disjoint manifolds separated by low den- sity). We exploit a novel algorithm for capturing manifold structure (high-order contractive auto-encoders) and we show how it builds a topological atlas of charts, each chart being characterized by the principal singular vectors of the Jacobian of a representation mapping. This representation learning algorithm can be stacked to yield a deep architecture, and we combine it with a domain knowledge-free version of the TangentProp algorithm to encourage the classiﬁer to be insensitive to local directions changes along the manifold. Record-breaking classiﬁcation results are obtained.