Metraipolis › archive › dc_ee3ce6155866

Learning Where to Steer: Noise-Space Geometry for Efficient Offline Multi-Objective Optimization with Generative Models — arXiv:2609.38920

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.38920
License (as archived)
CC0-1.0
Archived
2026-10-07T13:45:39.874Z
SHA-256 of the archived text
01f4cbc01b2e7ce5c31d5d9518a928296e2a29e4ed8eeaf2131dfa888ebb2346
Stable URL of this page
https://metraipolis.com/documento/dc_ee3ce6155866

How to cite

Learning Where to Steer: Noise-Space Geometry for Efficient Offline Multi-Objective Optimization with Generative Models — arXiv:2609.38920 — arXiv (en), https://arxiv.org/abs/2609.38920 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ee3ce6155866 (Metraipolis archive dc_ee3ce6155866, sha256 01f4cbc01b2e7ce5c31d5d9518a928296e2a29e4ed8eeaf2131dfa888ebb2346)
BibTeX
@misc{dcee3ce6155866,
  title = {Learning Where to Steer: Noise-Space Geometry for Efficient Offline Multi-Objective Optimization with Generative Models — arXiv:2609.38920},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.38920},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ee3ce6155866; sha256 01f4cbc01b2e7ce5c31d5d9518a928296e2a29e4ed8eeaf2131dfa888ebb2346},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.38920. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Offline multi-objective optimization (MOO) seeks solutions with better objective trade-offs using only a fixed dataset, without querying the objectives. Diffusion models trained on such data have emerged as a promising approach, but their samples are not inherently better than the data and must be steered toward the Pareto front. Existing methods guide or condition every sampling step. We instead act on the initial noise and leave the sampling process unchanged. Across Off-MOO-Bench, we observe that the objectives, as functions of the noise, are sensitive to only a few directions. We estimate these directions once per task via a Recursive Feature Machine using function values alone, and a small cache serves every trade-off, so each candidate costs one noise displacement and one ODE solve. We prove that this displacement increases the learned scalarized objective in expectation, and that sweeping trade-offs recovers the flow's attainable front up to proxy and steering errors. With additional guidance, for which we introduce novel data-adaptive and Pareto-aware operators, our method attains the best average hypervolume rank among generative methods on 47 tasks, at comparable or lower sampling cost. Steering alone outranks the best prior generative method at a fraction of its sampling cost.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.38920 — 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 2026-10-07T13:18:10.290Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).