Measuring the Symmetry–Data Exchange Rate A Controlled Measurement Under Exactly Known Symmetry

Ahmed M. Adly · 2026

Read Measuring the Symmetry–Data Exchange Rate A Controlled Measurement Under Exactly Known Symmetry by Ahmed M. Adly (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-2606-01090.txt.

Description

Adly measures how much data a symmetry prior is worth in this 2026 paper. Theory says a built in symmetry cuts the samples a model needs by the size of the symmetry group. The claim is widely cited but rarely measured as a scaling law. Confounds blur past tests: the prior, its fit, and its compute all shift at once.

His testbed is a controlled task with cyclic symmetry. A wrong group control, with the same orbit size and matched compute, does worse than no constraint at all. A mismatched prior harms, not merely fails to help. An augmentation baseline with test time orbit averaging matches the symmetric model bit for bit. The measured exchange rate is 1.28, near the predicted 1.0 in sign and scale in his main test. Wider bootstraps soften that match, and a finer replication is inconclusive. He supplies an estimator that cancels the shared difficulty confound.

Catalog record

CREATORAhmed M. Adly
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE82 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.01090

Browse the full library · About the Tower of Babel