Latent Space Factorisation and Manipulation via Matrix Subspace Projection
arXiv:1907.12385
Abstract
We tackle the problem disentangling the latent space of an autoencoder in order to separate labelled attribute information from other characteristic information. This then allows us to change selected attributes while preserving other information. Our method, matrix subspace projection, is much simpler than previous approaches to latent space factorisation, for example not requiring multiple discriminators or a careful weighting among their loss functions. Furthermore our new model can be applied to autoencoders as a plugin, and works across diverse domains such as images or text. We demonstrate the utility of our method for attribute manipulation in autoencoders trained across varied domains, using both human evaluation and automated methods. The quality of generation of our new model (e.g. reconstruction, conditional generation) is highly competitive to a number of strong baselines.
arXiv commentFinal camera ready version for ICML 2020
References in corpus (11)
- GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
- InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets
- Invertible Conditional GANs for image editing
- Evaluating the State-of-the-Art of End-to-End Natural Language Generation: The E2E NLG Challenge
- Toward Controlled Generation of Text
- Early Visual Concept Learning with Unsupervised Deep Learning
- i-RevNet: Deep Invertible Networks
- RelGAN: Multi-Domain Image-to-Image Translation via Relative Attributes
- Deep Nets: What have they ever done for Vision?
- Shaping the Narrative Arc: An Information-Theoretic Approach to Collaborative Dialogue
- Dirichlet Variational Autoencoder for Text Modeling