Speakers

Bayesian Deep Probabilistic Models

I_03_Seungjin Choi(Photo)
Seungjin Choi
POSTECH
14:40~15:20

Abstract :

Deep generative models are a rich class of models for density estimation which specify a generative process for observed data using a set of stochastic latent variables. Among those, of particular interest is the variational autoencoder (VAE) which is a deep directed generative model with continuous latent variables where a pair of probabilistic encoder (bottom-up) and decoder (top-down) is jointly learned by stochastic gradient variational Bayes (SGVB). In this talk, I begin with introducing our recent elaboration of Gaussian VAE, approximating the local covariance matrix of the decoder as an outer product of the principal direction at a position determined by a sample drawn from Gaussian distribution. I show that this model, referred to as VAE-ROC, better captures the data manifold, compared to the standard Gaussian VAE where independent multivariate Gaussian was used to model the decoder. Then, I introduce an extension of VAE-ROC to handle mixed ordinal and continuous data, where Gaussian copula is adopted to model the local dependency in mixed ordinal and continuous data, leading to “Gaussian copula variational autoencoder”.

 

Education

  • 1996 Ph.D in Electrical Engineering, University of Notre Dame, Indiana, USA
  • 1989 M.S. in Electrical Engineering, Seoul National University, Korea
  • 1987 B.S. in Electrical Engineering, Seoul National University, Korea

 

Professional Activities

  • 2016-Present Consulting Professor of Big Data Center of Shinhan Card
  • 2016-Present Consulting Professor of Ministry of the Interior
  • 2014-Present Director of POSTECH Machine Learning Center
  • 2014-Present Founding Chair of Korea Special Interest Group on Machine Learning

 

Research Interests

  • Bayesian nonparametric models
  • Bayesian learning and inference
  • Deep probabilistic models
  • Random projection
  • Matrix factorization
  • Multi-view learning
  • Multi-task, transfer learning
  • Zero-shot learning
  • Topic models
  • Anomaly detection
  • Recommendation systems

 

Awards & Honors

  • 2016-2017Senior Program Committee, AAAI
  • 2016 Reviewer, ECCV
  • 2016 Reviewer CVPR
  • 2016 Program Committee, ICML
  • 2015 Area Chair, NIPS
  • 2015 Reviewer, ICCV
  • 2012-2016 Reviewer, AISTATA
  • 2012 Program Co-Chair, SMC
  • 2011-2016 Reviewer, NIPS-2016, NIPS