Metraipolis › archive › dc_093af232d54f

Counterexamples to Local Reconstruction Gain as a Proxy for Final Fidelity in Residual Completion — arXiv:2609.34063

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.34063
License (as archived)
CC0-1.0
Archived
2026-10-07T21:54:09.257Z
SHA-256 of the archived text
48290ead8b6c690bf4b066fd95c00d1c9d19d324ee9d369404767c9a99b13564
Stable URL of this page
https://metraipolis.com/documento/dc_093af232d54f

How to cite

Counterexamples to Local Reconstruction Gain as a Proxy for Final Fidelity in Residual Completion — arXiv:2609.34063 — arXiv (en), https://arxiv.org/abs/2609.34063 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_093af232d54f (Metraipolis archive dc_093af232d54f, sha256 48290ead8b6c690bf4b066fd95c00d1c9d19d324ee9d369404767c9a99b13564)
BibTeX
@misc{dc093af232d54f,
  title = {Counterexamples to Local Reconstruction Gain as a Proxy for Final Fidelity in Residual Completion — arXiv:2609.34063},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.34063},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_093af232d54f; sha256 48290ead8b6c690bf4b066fd95c00d1c9d19d324ee9d369404767c9a99b13564},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.34063. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Residual completion augments query-aware sparse attention by estimating the contribution of tokens omitted from the exact sparse computation. We ask whether improving a layer's attention-output reconstruction on the same incoming Q/K/V and selected support necessarily improves the fidelity of the final model output. We study training-free RESA and learned Top-K+$\phi$ with frozen backbone language models. A prespecified single-layer screen yields two Qwen3-0.6B/Multi-LexSum interventions for which direct-runtime measurements show positive prespecified request-aggregate local reconstruction gain but worse final KL fidelity than the corresponding all-abstain Exact Top-K baseline on both discovery and prompt-token-disjoint holdout requests. Exact restoration at the same layer instead improves final fidelity, showing that the reversal is specific to approximate completion in these cases. In complementary multi-layer experiments, a task-independent local diagnostic often repairs the tested completion estimators, although the repaired models do not consistently outperform Exact Top-K. Together, these results show that better local reconstruction need not translate into better final-model fidelity.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.CL).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.34063 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T21:27:08.791Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).