Inter-Battery Topic Representation Learning

TL;DR

The paper introduces the Inter-Battery Topic Model (IBTM), a novel approach that learns structured, factorized latent representations from multi-view data, combining benefits of discriminative and generative models for improved classification.

Contribution

It proposes a new Bayesian model that factorizes latent variables for multi-view data, enabling effective feature selection and handling missing data, with demonstrated state-of-the-art results.

Findings

Achieves state-of-the-art classification accuracy with compact representations

Effectively models shared and view-specific topics in multi-view data

Provides robust inference and efficient learning in multi-modality settings

Abstract

In this paper, we present the Inter-Battery Topic Model (IBTM). Our approach extends traditional topic models by learning a factorized latent variable representation. The structured representation leads to a model that marries benefits traditionally associated with a discriminative approach, such as feature selection, with those of a generative model, such as principled regularization and ability to handle missing data. The factorization is provided by representing data in terms of aligned pairs of observations as different views. This provides means for selecting a representation that separately models topics that exist in both views from the topics that are unique to a single view. This structured consolidation allows for efficient and robust inference and provides a compact and efficient representation. Learning is performed in a Bayesian fashion by maximizing a rigorous bound on the log-likelihood. Firstly, we illustrate the benefits of the model on a synthetic dataset,. The model is then evaluated in both uni- and multi-modality settings on two different classification tasks with off-the-shelf convolutional neural network (CNN) features which generate state-of-the-art results with extremely compact representations.