Dr. Ramses Sanchez

I’m a theoretical physicist with a background on statistical mechanics and out-of-equilibrium dynamics. My research interests mainly center on developing (deep) generative models to study dynamical systems and different aspects of natural language.

Problems I'm working on:

1. Representation learning for natural language understanding

   We develop models in which sentences are generated conditioned on (latent) stochastic processes, which are in turn conditioned on a global semantic object (à la Jackendoff). One of our goals is automatic commonsense knowledge-based completion;
   We use information theory to improve conditional generation in stochastic AE models of language and allow for unsupervised “topic modelling” at the sentence level. We plan to use the learned representations for unsupervised abstractive summarization.

2. Neural-based variational inference for coupled Markov Jump processes.

3. Infinite-width and infinite-dept limits of stochastic autoencoders.