Postdoc position offer - Institut 3IA Côte d'Azur - Deep Latent Variable Models for the Analysis of Massive Evolving Heterogenous Data

A first goal of this Ph.D. will be to propose a generative DLVM model specifically designed for massive evolving heterogenous
data. Regarding the problem of model selection in this context, some preliminary studies we performed have highlighted the surprising fact that the evidence lower bound of the fitted model may be used as a model selection criterion in some extent. This strongly suggests revisiting
the study of these latent variable models with a Bayesian point of view and to understand how this evidence lower bound integrate implicit priors on the latent variables. Having a clear understanding of this point will offer an elegant and powerful tool for picking the appropriate
model (latent dimensions, network architecture, network sparsity, … ) for the data at hand. The proposed methodologies will be then applied to real-world situations in either Medicine (Pharmacovigilance, omics-based clinical discovery, …) or Digital Humanities (History, Archeology, …).

Expected skills: The candidate should have a graduate degree (Master 2 degree). Him/her
scholar background should include:
• statistical/machine learning, statistical inference, clustering, classification
• deep learning, variational auto-encoder, back-propagation,
• knowledge of R (main programming language), Python and C++.

Application: Application files should contain a resumé, an application letter and grade
records of the 2 last years (M1 & M2). Applications should be sent by email to
charles.bouveyron@inria.fr.