Disentangling Factors of Variation via Generative Entangling

TL;DR

This paper introduces a generative model based on spike-and-slab restricted Boltzmann machines that learns to disentangle underlying factors of variation in data without supervision, demonstrated on facial expression classification.

Contribution

The paper presents a novel generative model with higher-order interactions for unsupervised disentangling of data factors, extending previous models with a new approach.

Findings

Successfully disentangles factors of variation in data

Achieves competitive results in facial expression classification

Operates without supervised labels for latent factors

Abstract

Here we propose a novel model family with the objective of learning to disentangle the factors of variation in data. Our approach is based on the spike-and-slab restricted Boltzmann machine which we generalize to include higher-order interactions among multiple latent variables. Seen from a generative perspective, the multiplicative interactions emulates the entangling of factors of variation. Inference in the model can be seen as disentangling these generative factors. Unlike previous attempts at disentangling latent factors, the proposed model is trained using no supervised information regarding the latent factors. We apply our model to the task of facial expression classification.