The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement
Part of the Lecture Notes in Computer Science book series (LNCS, volume 12351)
| Main Authors: | , , , , |
|---|---|
| Other Authors: | |
| Format: | Book |
| Language: | English |
| Published: |
Springer International Publishing
2021
|
| Online Access: | https://hdl.handle.net/1721.1/130353 |
| _version_ | 1826211676353986560 |
|---|---|
| author | Peebles, William Peebles, John Zhu, Jun-Yan Efros, Alexei Torralba, Antonio |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Peebles, William Peebles, John Zhu, Jun-Yan Efros, Alexei Torralba, Antonio |
| author_sort | Peebles, William |
| collection | MIT |
| description | Part of the Lecture Notes in Computer Science book series (LNCS, volume 12351) |
| first_indexed | 2024-09-23T15:09:45Z |
| format | Book |
| id | mit-1721.1/130353 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T15:09:45Z |
| publishDate | 2021 |
| publisher | Springer International Publishing |
| record_format | dspace |
| spelling | mit-1721.1/1303532022-10-02T01:00:30Z The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement Peebles, William Peebles, John Zhu, Jun-Yan Efros, Alexei Torralba, Antonio Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Part of the Lecture Notes in Computer Science book series (LNCS, volume 12351) Existing disentanglement methods for deep generative models rely on hand-picked priors and complex encoder-based architectures. In this paper, we propose the Hessian Penalty, a simple regularization term that encourages the Hessian of a generative model with respect to its input to be diagonal. We introduce a model-agnostic, unbiased stochastic approximation of this term based on Hutchinson’s estimator to compute it efficiently during training. Our method can be applied to a wide range of deep generators with just a few lines of code. We show that training with the Hessian Penalty often causes axis-aligned disentanglement to emerge in latent space when applied to ProGAN on several datasets. Additionally, we use our regularization term to identify interpretable directions in BigGAN’s latent space in an unsupervised fashion. Finally, we provide empirical evidence that the Hessian Penalty encourages substantial shrinkage when applied to over-parameterized latent spaces. We encourage readers to view videos of our disentanglement results at www.wpeebles.com/hessian-penalty, and code at https://github.com/wpeebles/hessian_penalty. 2021-04-02T15:45:58Z 2021-04-02T15:45:58Z 2020-11 2021-01-28T16:07:58Z Book http://purl.org/eprint/type/ConferencePaper 9783030585389 9783030585396 0302-9743 1611-3349 https://hdl.handle.net/1721.1/130353 Peebles, William et al. "The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement." ECCV: European Conference on Computer Vision, Lecture Notes in Computer Science, 12351, Springer International Publishing, 2020, 581-597. © 2020, Springer Nature en http://dx.doi.org/10.1007/978-3-030-58539-6_35 Lecture Notes in Computer Science Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf Springer International Publishing arXiv |
| spellingShingle | Peebles, William Peebles, John Zhu, Jun-Yan Efros, Alexei Torralba, Antonio The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement |
| title | The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement |
| title_full | The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement |
| title_fullStr | The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement |
| title_full_unstemmed | The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement |
| title_short | The Hessian Penalty: A Weak Prior for Unsupervised Disentanglement |
| title_sort | hessian penalty a weak prior for unsupervised disentanglement |
| url | https://hdl.handle.net/1721.1/130353 |
| work_keys_str_mv | AT peebleswilliam thehessianpenaltyaweakpriorforunsuperviseddisentanglement AT peeblesjohn thehessianpenaltyaweakpriorforunsuperviseddisentanglement AT zhujunyan thehessianpenaltyaweakpriorforunsuperviseddisentanglement AT efrosalexei thehessianpenaltyaweakpriorforunsuperviseddisentanglement AT torralbaantonio thehessianpenaltyaweakpriorforunsuperviseddisentanglement AT peebleswilliam hessianpenaltyaweakpriorforunsuperviseddisentanglement AT peeblesjohn hessianpenaltyaweakpriorforunsuperviseddisentanglement AT zhujunyan hessianpenaltyaweakpriorforunsuperviseddisentanglement AT efrosalexei hessianpenaltyaweakpriorforunsuperviseddisentanglement AT torralbaantonio hessianpenaltyaweakpriorforunsuperviseddisentanglement |