analiza aiqbench
This research proposes an unsupervised objective to directly learn global geometric structures in representation spaces by enforcing commutativity, time coherence, and common-reference properties. By shifting from a reconstruction-based "map-first" approach to one that explicitly optimizes these structural constraints, the study aims to prevent small approximation errors from disrupting the global organization of data manifolds. Theoretical results and experiments on controlled manifolds validate that this method effectively recovers tangent spaces and maintains robust global structure even when standard autoencoder baselines fail.
Discovering Data Manifold Geometry through Geometric Properties
arXiv:2602.02611v2 Announce Type: replace-cross Abstract: A prevailing paradigm in modern representation learning is the map-first approach, in which a representation map is learned from reconstruction, embedding, or task objectives. At the optimum, when the learned map accurately recovers a global coordinate chart, it should exhibit three structural properties whose geometric meaning can be illustrated through a face-editing example: Commutativity requires that changing pose and then expression gives the same result as applying them in the reverse order; Time Coherence requires that the same variation along one coordinate induces the same expression change across faces; Common-Reference requires that all faces are organized relative to a common reference face. However, small approximation errors in the learned map need not translate into small errors in these structural properties, and can therefore disrupt the global organization of the representation. Based on this observation, we consider the converse of the map-first formulation and ask whether a global representation can instead emerge by directly learning these properties. We represent variations along individual coordinates through vector fields defined in the ambient space and introduce a non-contraction condition preventing one transformation from destroying directions associated with the others. We derive an unsupervised objective that learns these structural properties and establish theoretical results connecting its minimization to tangent-space recovery. Experiments on controlled manifolds validate the predicted tangent-space recovery and global structure, while an autoencoder baseline shows that small map-first errors can still produce substantial violations of the targeted properties.
Abstract:A prevailing paradigm in modern representation learning is the map-first approach, in which a representation map is learned from reconstruction, embedding, or task objectives. At the optimum, when the learned map accurately recovers a global coordinate chart, it should exhibit three structural properties whose geometric meaning can be illustrated through a face-editing example: Commutativity requires that changing pose and then expression gives the same result as applying them in the reverse order; Time Coherence requires that the same variation along one coordinate induces the same expression change across faces; Common-Reference requires that all faces are organized relative to a common reference face. However, small approximation errors in the learned map need not translate into small errors in these structural properties, and can therefore disrupt the global organization of the representation. Based on this observation, we consider the converse of the map-first formulation and ask whether a global representation can instead emerge by directly learning these properties. We represent variations along individual coordinates through vector fields defined in the ambient space and introduce a non-contraction condition preventing one transformation from destroying directions associated with the others. We derive an unsupervised objective that learns these structural properties and establish theoretical results connecting its minimization to tangent-space recovery. Experiments on controlled manifolds validate the predicted tangent-space recovery and global structure, while an autoencoder baseline shows that small map-first errors can still produce substantial violations of the targeted properties.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2602.02611 [cs.LG]
(or arXiv:2602.02611v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2602.02611 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 330 cuvinte
Subiecte conexe
Entanglement entropy reveals hidden connections between dimensions
Quantum Zeitgeist · 12 oct. 2026
Variational autoencoders with latent high-dimensional steady geometric flows for dynamics
arXiv · 17 iun. 2026
Entropic Optimal Transport Eigenmaps for Nonlinear Alignment and Joint Embedding of High-Dimensional Datasets
arXiv · 9 iun. 2026
PIRAMID: Progress and Plans
LessWrong · 27 iul. 2026