Adaptivity Under Realizability Constraints: Comparing In-Context and Agentic Learning

Anastasis Kratsios, A. Martina Neuman, Philipp Petersen · 2026

Read Adaptivity Under Realizability Constraints: Comparing In-Context and Agentic Learning by Anastasis Kratsios, A. Martina Neuman, Philipp Petersen (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-2605-04995.txt.

Description

Anastasis Kratsios, Martina Neuman, and Philipp Petersen compare learning from fixed questions with learning that adapts its questions as it goes, for approximating families of tasks evenly. They study two regimes: one where any functions are allowed, and one where the operations must be built from ReLU neural networks, the building blocks of real systems. Adaptivity, they show, never hurts in either regime.

The advantage itself, however, can change when the network constraint is imposed. Four explicit task families witness four outcomes: no advantage, an advantage that survives, one that appears only under the constraint, and one that vanishes. Representational limits, they conclude, interact deeply with adaptivity.

Catalog record

CREATORAnastasis Kratsios, A. Martina Neuman, Philipp Petersen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.04995

Browse the full library · About the Tower of Babel