Metraipolis › archive › dc_fd0f06015f6f

Understanding Curriculum Learning in Large Language Models via Cross-Difficulty Optimization Dynamics — arXiv:2608.17268

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.17268
License (as archived)
CC0-1.0
Archived
2026-10-18T14:29:23.576Z
SHA-256 of the archived text
50cd2e64c27aadd46a749e897cc314efbca123c9b6497fa992abe720f4d2320e
Stable URL of this page
https://metraipolis.com/documento/dc_fd0f06015f6f

How to cite

Understanding Curriculum Learning in Large Language Models via Cross-Difficulty Optimization Dynamics — arXiv:2608.17268 — arXiv (en), https://arxiv.org/abs/2608.17268 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fd0f06015f6f (Metraipolis archive dc_fd0f06015f6f, sha256 50cd2e64c27aadd46a749e897cc314efbca123c9b6497fa992abe720f4d2320e)
BibTeX
@misc{dcfd0f06015f6f,
  title = {Understanding Curriculum Learning in Large Language Models via Cross-Difficulty Optimization Dynamics — arXiv:2608.17268},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.17268},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fd0f06015f6f; sha256 50cd2e64c27aadd46a749e897cc314efbca123c9b6497fa992abe720f4d2320e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.17268. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Curriculum learning has been widely adopted in the post-training of large language models by organizing training data from easy to hard. However, its effectiveness varies substantially across reasoning tasks, suggesting that no single curriculum is universally optimal and raising a fundamental question: what determines when curriculum learning works? In this paper, we answer this question by analyzing the optimization dynamics induced by different curriculum schedules. We show that the transfer relationship between different difficulty levels characterizes the optimization dynamics induced by curriculum learning, which in turn explains the effectiveness of different curriculum schedules, and formalize this relationship as Relative Transfer, a principled measure of cross-difficulty knowledge transfer. Based on this measurement, we derive Transfer-aware Dynamic Curriculum Sampling (TDCS), which dynamically adjusts the sampling distribution according to the estimated transfer relationship throughout training. Extensive experiments on multiple reasoning benchmarks demonstrate that TDCS consistently outperforms representative scheduling strategies across different tasks, model scales, and training paradigms. More importantly, our work provides a unified optimization-based explanation of curriculum learning through cross-difficulty transfer.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-19 in cs.AI).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2608.17268 — 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 2026-10-18T14:27:54.406Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-19?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).