Learning Disentangled Joint Continuous and Discrete Representations
This addresses the need for interpretable representations in machine learning, though it is incremental as it builds on variational autoencoders.
The paper tackles the problem of learning disentangled joint continuous and discrete representations in an unsupervised manner, achieving automatic discovery of these factors and outperforming current methods when discrete factors are prominent.
We present a framework for learning disentangled and interpretable jointly continuous and discrete representations in an unsupervised manner. By augmenting the continuous latent distribution of variational autoencoders with a relaxed discrete distribution and controlling the amount of information encoded in each latent unit, we show how continuous and categorical factors of variation can be discovered automatically from data. Experiments show that the framework disentangles continuous and discrete generative factors on various datasets and outperforms current disentangling methods when a discrete generative factor is prominent.