A Function-Space Stability Boundary for Generalization in Interpolating Learning Systems

Ronald Katende · 2026

Read A Function-Space Stability Boundary for Generalization in Interpolating Learning Systems by Ronald Katende (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-03514.txt.

Description

Modern learning systems often fit their training data exactly yet still perform well on new data, and stability is one offered account. Ronald Katende examines when that account actually holds in a 2026 paper. He models training as a trajectory through function space and measures how sensitive it is when a single training sample changes.

Katende proposes a contractive condition on how effects spread, and unrolls it into a stability certificate. A small certificate guarantees stability-based generalization, but he also proves some low-risk exact-fit regimes cannot satisfy the condition at all. In experiments, faster certificate growth goes with worse generalization across optimizers, step sizes, and changed datasets. The framework maps where stability explains success and where other causes must.

Catalog record

CREATORRonald Katende
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-03514

Browse the full library · About the Tower of Babel