Disentanglement by means of action-induced representations
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912884647264256 |
|---|---|
| author | Muñoz-Gil, Gorka Nautrup, Hendrik Poulsen Majumder, Arunava de Schoulepnikoff, Paulin Fürrutter, Florian Krumm, Marius Briegel, Hans J. |
| author_facet | Muñoz-Gil, Gorka Nautrup, Hendrik Poulsen Majumder, Arunava de Schoulepnikoff, Paulin Fürrutter, Florian Krumm, Marius Briegel, Hans J. |
| contents | Learning interpretable representations with variational autoencoders (VAEs) is a major goal of representation learning. The main challenge lies in obtaining disentangled representations, where each latent dimension corresponds to a distinct generative factor. This difficulty is fundamentally tied to the inability to perform nonlinear independent component analysis. Here, we introduce the framework of action-induced representations (AIRs) which models representations of physical systems given experiments (or actions) that can be performed on them. We show that, in this framework, we can provably disentangle degrees of freedom w.r.t. their action dependence. We further introduce a variational AIR architecture (VAIR) that can extract AIRs and therefore achieve provable disentanglement where standard VAEs fail. Beyond state representation, VAIR also captures the action dependence of the underlying generative factors, directly linking experiments to the degrees of freedom they influence. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_06741 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Disentanglement by means of action-induced representations Muñoz-Gil, Gorka Nautrup, Hendrik Poulsen Majumder, Arunava de Schoulepnikoff, Paulin Fürrutter, Florian Krumm, Marius Briegel, Hans J. Machine Learning Data Analysis, Statistics and Probability Quantum Physics Learning interpretable representations with variational autoencoders (VAEs) is a major goal of representation learning. The main challenge lies in obtaining disentangled representations, where each latent dimension corresponds to a distinct generative factor. This difficulty is fundamentally tied to the inability to perform nonlinear independent component analysis. Here, we introduce the framework of action-induced representations (AIRs) which models representations of physical systems given experiments (or actions) that can be performed on them. We show that, in this framework, we can provably disentangle degrees of freedom w.r.t. their action dependence. We further introduce a variational AIR architecture (VAIR) that can extract AIRs and therefore achieve provable disentanglement where standard VAEs fail. Beyond state representation, VAIR also captures the action dependence of the underlying generative factors, directly linking experiments to the degrees of freedom they influence. |
| title | Disentanglement by means of action-induced representations |
| topic | Machine Learning Data Analysis, Statistics and Probability Quantum Physics |
| url | https://arxiv.org/abs/2602.06741 |