Metraipolis › archive › dc_10facddc53b9

Intervention-Certified Amplification: A Validity Gate for $σ$-Weighted Post-Training — arXiv:2610.33882

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.33882
License (as archived)
CC0-1.0
Archived
2026-10-27T08:00:33.108Z
SHA-256 of the archived text
207de7861eaa82c6cc8924697e2bee97c837c4acf1ce8a54e8313293698adce2
Stable URL of this page
https://metraipolis.com/documento/dc_10facddc53b9

How to cite

Intervention-Certified Amplification: A Validity Gate for $σ$-Weighted Post-Training — arXiv:2610.33882 — arXiv (en), https://arxiv.org/abs/2610.33882 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_10facddc53b9 (Metraipolis archive dc_10facddc53b9, sha256 207de7861eaa82c6cc8924697e2bee97c837c4acf1ce8a54e8313293698adce2)
BibTeX
@misc{dc10facddc53b9,
  title = {Intervention-Certified Amplification: A Validity Gate for $σ$-Weighted Post-Training — arXiv:2610.33882},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.33882},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_10facddc53b9; sha256 207de7861eaa82c6cc8924697e2bee97c837c4acf1ce8a54e8313293698adce2},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.33882. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Hybrid post-training often mixes supervised fine-tuning (SFT) and reinforcement learning (RL), using training-side signals to set each sample's balance. Such signals characterize the anchor data and policy, but not whether the emphasized behavior helps in the target task and evaluation setting. This mismatch can reactivate and amplify pretrained behavior even after its literal form is removed from supervision. We demonstrate this with DIAA (Dual-Implicit Adversarial Amplification), a per-sample $\sigma$-weighted SFT/RL hybrid. With all literal Python fences removed from its anchor, DIAA emits unexecuted Tool-Integrated Reasoning (TIR) on $29%$ of MATH-500 prompts, three times vanilla GRPO. Paired interventions show $10.6$ percentage points of accuracy recovery when TIR is suppressed and comparable contributions from SFT imitation and $\sigma$-amplification; a student trained on a $0%$-TIR teacher anchor emits it at $45.8%$. Intervention-Certified Amplification (ICA) gates $\sigma$ with paired-rollout target evidence. Frozen and online ICA reduce TIR to $13.8%$ and $11.2%$, halve residual drag, improve accuracy from $47.5%$ to $48.7%$ and $49.3%$, and transfer to held-out GSM8K without retuning.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.33882 — 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 2026-10-27T07:55:30.861Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).