Metraipolis › archive › dc_cbff021c1c32

Strategically Diverse Sampling for Self-Training — arXiv:2609.31571

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.31571
License (as archived)
CC0-1.0
Archived
2026-10-08T02:04:11.275Z
SHA-256 of the archived text
b382322f83d6c4d6d30965e0cdf4c4a92b247a95f0fbdb4e70655d7dce4bc338
Stable URL of this page
https://metraipolis.com/documento/dc_cbff021c1c32

How to cite

Strategically Diverse Sampling for Self-Training — arXiv:2609.31571 — arXiv (en), https://arxiv.org/abs/2609.31571 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cbff021c1c32 (Metraipolis archive dc_cbff021c1c32, sha256 b382322f83d6c4d6d30965e0cdf4c4a92b247a95f0fbdb4e70655d7dce4bc338)
BibTeX
@misc{dccbff021c1c32,
  title = {Strategically Diverse Sampling for Self-Training — arXiv:2609.31571},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.31571},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cbff021c1c32; sha256 b382322f83d6c4d6d30965e0cdf4c4a92b247a95f0fbdb4e70655d7dce4bc338},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.31571. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Many LLM training and inference methods, including RL and test-time scaling, depend on repeated sampling, but benefit only when the responses meaningfully differ. Self-training faces the same challenge: training data is typically constructed by sampling IID responses and filtering primarily for correctness, thereby overrepresenting strategies a model already favours. We investigate strategic diversity, or substantive variation among approaches to a problem, as an alternative principle for constructing self-training data. We generate strategically diverse data with two sampling methods: GROOT, a new method which constructs a hierarchical tree of approaches and samples distinct paths, and Verbalized Sampling (VS), adapted to produce an unstructured set of approaches. Across competitive programming and Next-Chapter Prediction domains, models trained on strategically sampled data outperform IID-trained counterparts on difficult tasks and provide strong initializations for RL and test-time scaling. Most strikingly, self-training on strategically diverse but incorrect traces from Qwen3-4B outperforms IID distillation from a 235B teacher. These results challenge prevailing assumptions about what makes useful self-training data and show that diversity of approaches can matter more than correctness or teacher scale.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-28 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.31571 — 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 2026-10-08T01:38:49.469Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-28?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).