Variational Interaction Information Maximization for Cross-domain Disentanglement
This work addresses the problem of learning disentangled representations for domain transfer and semantic distance measurement, which is relevant for researchers and practitioners in cross-domain machine learning.
This paper tackles the problem of cross-domain disentanglement, aiming to learn representations separated into domain-invariant and domain-specific components. The authors propose a generative model called Interaction Information Auto-Encoder (IIAE) that achieves state-of-the-art performance in zero-shot sketch-based image retrieval without external knowledge.
Cross-domain disentanglement is the problem of learning representations partitioned into domain-invariant and domain-specific representations, which is a key to successful domain transfer or measuring semantic distance between two domains. Grounded in information theory, we cast the simultaneous learning of domain-invariant and domain-specific representations as a joint objective of multiple information constraints, which does not require adversarial training or gradient reversal layers. We derive a tractable bound of the objective and propose a generative model named Interaction Information Auto-Encoder (IIAE). Our approach reveals insights on the desirable representation for cross-domain disentanglement and its connection to Variational Auto-Encoder (VAE). We demonstrate the validity of our model in the image-to-image translation and the cross-domain retrieval tasks. We further show that our model achieves the state-of-the-art performance in the zero-shot sketch based image retrieval task, even without external knowledge. Our implementation is publicly available at: https://github.com/gr8joo/IIAE