Manifold Relevance Determination

TL;DR

This paper introduces a Bayesian latent variable model that learns shared and private representations from multi-view data, capable of modeling high-dimensional spaces and generating new data, with automatic latent dimension estimation.

Contribution

It presents a novel fully Bayesian approach with relaxed sharing constraints, enabling efficient modeling of complex, high-dimensional data and automatic latent space dimension estimation.

Findings

Successfully models high-dimensional image data with thousands of pixels.

Generates novel images by sampling from learned latent spaces.

Effectively predicts human pose in ambiguous scenarios.

Abstract

In this paper we present a fully Bayesian latent variable model which exploits conditional nonlinear(in)-dependence structures to learn an efficient latent representation. The latent space is factorized to represent shared and private information from multiple views of the data. In contrast to previous approaches, we introduce a relaxation to the discrete segmentation and allow for a "softly" shared latent space. Further, Bayesian techniques allow us to automatically estimate the dimensionality of the latent spaces. The model is capable of capturing structure underlying extremely high dimensional spaces. This is illustrated by modelling unprocessed images with tenths of thousands of pixels. This also allows us to directly generate novel images from the trained model by sampling from the discovered latent spaces. We also demonstrate the model by prediction of human pose in an ambiguous setting. Our Bayesian framework allows us to perform disambiguation in a principled manner by including latent space priors which incorporate the dynamic nature of the data.