Learning Transfers: Kan Extensions of Topological Invariants of Learned Representations
Abstract
When a learned representation is reused for a new task, which of its structure should survive? A task is a small category, a change of task is a functor J, and the invariants of a trained representation form a functor F. The left Kan extension LanJF is the invariant that the source and the change of task force on the target: by its universal property, every comparison of the source with an observed target invariant factors through it. The transfer discrepancy is the largest distance, over the target objects, between this extension and the observed invariant. For persistent homology the extension is a cokernel over comma categories, and a functorial version of the discrepancy, which also compares the maps between target objects, is $1$-Lipschitz under interleavings. On finite samples the discrepancy is a bottleneck distance, computed without approximation. For data glued from two pieces whose exclusive parts lie farther apart than the filtration scale, a Mayer-Vietoris sequence identifies what extending homology loses, while extending chains loses nothing. For a merge of labelled point clouds, the degree-zero discrepancy vanishes if and only if some $1$-Lipschitz classifier has a margin above the filtration scale at every sample point. In experiments on sampled manifolds, trained encoders and pretrained image models, every exact prediction holds on every sample. The discrepancy of truncated invariants recovers a hidden merge of separated parts far above chance and tells the intended target of a merge from a permuted control, but across pretrained models it does not track the accuracy of a linear probe.
Related
- Learning Topological Representations of Protein Structure and Dynamics
- Computing the Iwasawa μ-Invariants for Elliptic Curves over ℤ2-Extensions
- Ramification invariants of pure extensions
- Non-admissible smooth irreducible representations of GL2(F) for totally ramified extensions F/ℚp
- Flag Varieties of General Quiver Representations: Characteristic Classes and Semi-Invariants
BibTeX
@misc{melodia2026learning,
title = {{Learning Transfers: Kan Extensions of Topological Invariants of Learned Representations}},
author = {Luciano Melodia},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2606.07627},
eprint = {2606.07627},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.07627},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Learning Transfers: Kan Extensions of Topological Invariants of Learned Representations
AU - Luciano Melodia
PY - 2026
DA - 2026/10/09
PB - arXiv
DO - 10.48550/arxiv.2606.07627
UR - https://arxiv.org/abs/2606.07627
AB - When a learned representation is reused for a new task, which of its structure should survive? A task is a small category, a change of task is a functor J, and the invariants of a trained representation form a functor F. The left Kan extension Lan_JF is the invariant that the source and the change of task force on the target: by its universal property, every comparison of the source with an observed target invariant factors through it. The transfer discrepancy is the largest distance, over the target objects, between this extension and the observed invariant. For persistent homology the extension is a cokernel over comma categories, and a functorial version of the discrepancy, which also compares the maps between target objects, is $1$-Lipschitz under interleavings. On finite samples the discrepancy is a bottleneck distance, computed without approximation. For data glued from two pieces whose exclusive parts lie farther apart than the filtration scale, a Mayer-Vietoris sequence identifies what extending homology loses, while extending chains loses nothing. For a merge of labelled point clouds, the degree-zero discrepancy vanishes if and only if some $1$-Lipschitz classifier has a margin above the filtration scale at every sample point. In experiments on sampled manifolds, trained encoders and pretrained image models, every exact prediction holds on every sample. The discrepancy of truncated invariants recovers a hidden merge of separated parts far above chance and tells the intended target of a merge from a permuted control, but across pretrained models it does not track the accuracy of a linear probe.
ER -