What We Don't C: Manifold Disentanglement for Structured Discovery
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911505447911424 |
|---|---|
| author | Rogers, Brian Bowles, Micah Lintott, Chris J. Croft, Steve King, Oliver N. F. Ray, James Kostas |
| author_facet | Rogers, Brian Bowles, Micah Lintott, Chris J. Croft, Steve King, Oliver N. F. Ray, James Kostas |
| contents | Accessing information in learned representations is critical for annotation, discovery, and data filtering in disciplines where high-dimensional datasets are common. We introduce What We Don't C, a novel approach based on latent flow matching that disentangles latent subspaces by explicitly removing information included in conditional guidance, resulting in meaningful residual representations. This allows factors of variation which have not already been captured in conditioning to become more readily available. We show how guidance in the flow path necessarily represses the information from the guiding, conditioning variables. Our results highlight this approach as a simple yet powerful mechanism for analyzing, controlling, and repurposing latent representations, providing a pathway toward using generative models to explore what we don't capture, consider, or catalog. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_09433 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | What We Don't C: Manifold Disentanglement for Structured Discovery Rogers, Brian Bowles, Micah Lintott, Chris J. Croft, Steve King, Oliver N. F. Ray, James Kostas Artificial Intelligence Accessing information in learned representations is critical for annotation, discovery, and data filtering in disciplines where high-dimensional datasets are common. We introduce What We Don't C, a novel approach based on latent flow matching that disentangles latent subspaces by explicitly removing information included in conditional guidance, resulting in meaningful residual representations. This allows factors of variation which have not already been captured in conditioning to become more readily available. We show how guidance in the flow path necessarily represses the information from the guiding, conditioning variables. Our results highlight this approach as a simple yet powerful mechanism for analyzing, controlling, and repurposing latent representations, providing a pathway toward using generative models to explore what we don't capture, consider, or catalog. |
| title | What We Don't C: Manifold Disentanglement for Structured Discovery |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2511.09433 |