Demystifying Inter-Class Disentanglement

By Aviv Gabbay, Yedid Hoshen. 2019

65 citations recorded by the source corpus.

Learning to disentangle the hidden factors of variations within a set of observations is a key task for artificial intelligence. We present a unified formulation for class and content disentanglement and use it to illustrate the limitations of current methods. We therefore introduce LORD, a novel method based on Latent Optimization for Representation Disentanglement. We find that latent optimization, along with an asymmetric noise regularization, is superior to amortized inference for achieving disentangled representations. In extensive experiments, our method is shown to achieve better disentanglement performance than both adversarial and non-adversarial methods that use the same level of supervision. We further introduce a clustering-based approach for extending our method for settings that exhibit in-class variation with promising results on the task of domain translation.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Stefano Ermon
2. Ben Poole
3. Abhishek Kumar
4. Sergey Levine
5. Daniel Cohen-Or
6. Yang Song
7. Rui Shu
8. Jiaming Song

In Computer Vision and Pattern Recognition

1. Daniel Cohen-Or
2. Luc Van Gool
3. Dacheng Tao
4. Ming-Hsuan Yang
5. Jitendra Malik
6. Stefano Ermon
7. Andrea Vedaldi
8. Trevor Darrell

Citations in the corpus

• Weakly Supervised Disentanglement with Guarantees (2019, methodology): “To build robust disentangled representation learning methods that do not require large amounts of supervised data, recent work has turned to forms of weak supervision (Chen & Batmanghelich, 2019; Gabbay & Hoshen, 2019).”
• Face identity disentanglement via latent space mapping (2020, methodology): “Unlike other methods [Bouchacourt et al. 2018; Denton et al. 2017; Gabbay and Hoshen 2019; Hadad et al. 2018], our training data does not contain examples of the same person twice, nor does it have any indication or labeling regarding the person’s identity.”
• Disentangling in Latent Space by Harnessing a Pretrained Generator (2020, methodology): “Unlike other methods [18, 7, 22, 16], our training data does not contain examples of the same person twice, nor does it have any indication or labeling regarding the person’s identity.”