We propose to exploit {\em reconstruction} as a layer-local training signal for deep learning.
Reconstructions can be propagated in a form of target propagation playing a role similar to …

We introduce a novel training principle for generative probabilistic models that is an
alternative to maximum likelihood. The proposed Generative Stochastic Networks (GSNs) …

We extend generative stochastic networks to supervised learning of representations. In
particular, we introduce a hybrid training objective considering a generative and …

In this paper, we use deep representation learning for model-based single-channel source
separation (SCSS) and artificial bandwidth extension (ABE). Both tasks are ill-posed and …

Recent work by Bengio et al.(2013) proposes a sampling procedure for denoising
autoencoders which involves learning the transition operator of a Markov chain. The …

Single channel source separation (SCSS) is ill-posed and thus challenging. In this paper,
we apply general stochastic networks (GSNs)–a deep neural network architecture–to SCSS …

Autoregressive models factorize a multivariate joint probability distribution into a product of
one-dimensional conditional distributions. The variables are assigned an ordering, and the …

Recent work by Bengio et al.(2013) proposes a sampling procedure for denoising
autoencoders which involves learning the transition operator of a Markov chain. The …

Recent discoveries in the field of biology have transformed it into a data-rich domain. This
has invited multiple machine learning applications, and in particular, deep learning a set of …

The goal of this thesis is to present a body of work that serves as my modest contribution to
humanity's quest to understand intelligence and to implement intelligent systems. This is a …