Understanding Curriculum Learning in Large Language Models via Cross-Difficulty Optimization Dynamics

• URL: http://arxiv.org/abs/2608.17268v1
• Date: Tue, 18 Aug 2026 01:51:20 GMT
• Title: Understanding Curriculum Learning in Large Language Models via Cross-Difficulty Optimization Dynamics
• Abstract summary: We show that the transfer relationship between different difficulty levels characterizes the optimization dynamics induced by curriculum learning.<n>Our work provides a unified optimization-based explanation of curriculum learning through cross-difficulty transfer.
• Score: 5.2200534546273465
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• SPARD: Self-Paced Curriculum for RL Alignment via Integrating Reward Dynamics and Data Utility [71.76390626651254]
  We propose SPARD, a framework that establishes an automated, self-paced curriculum by perceiving learning progress to adjust dynamically multi-objective reward weights and data importance.<n>Extensive experiments across multiple benchmarks demonstrate that SPARD significantly enhances model capabilities across all domains.
  arXiv  Detail & Related papers  (2026-04-09T05:37:22Z)
• Data Organization Matters in Multimodal Instruction Tuning: A Controlled Study of Capability Trade-offs [0.0]
  We study whether data organization affects the trade-off among general understanding, structured reasoning, and fine-grained OCR/document understanding.<n>We compare four strategies: direct mixture, curriculum training, balanced sampling, and reverse curriculum.
  arXiv  Detail & Related papers  (2026-03-29T15:54:06Z)
• Level Up: Defining and Exploiting Transitional Problems for Curriculum Learning [11.76670314650204]
  We introduce a novel method for measuring the difficulty of individual problem instances directly relative to the ability of a given model.<n>We find that training on a curriculum that "levels up" from easier to harder transitional problems most efficiently improves a model to the next tier of competence.
  arXiv  Detail & Related papers  (2026-03-14T05:17:45Z)
• Scaling and Transferability of Annealing Strategies in Large Language Model Training [59.443651879173025]
  We refine a predictive framework for optimizing annealing strategies under the Warmup-Steady-Decay (WSD) scheduler.<n>Our improved framework incorporates training steps, maximum learning rate, and annealing behavior, enabling more efficient optimization of learning rate schedules.<n>We validate our findings on extensive experiments using both Dense and Mixture-of-Experts (MoE) models.
  arXiv  Detail & Related papers  (2025-12-05T16:38:33Z)
• What Makes a Good Curriculum? Disentangling the Effects of Data Ordering on LLM Mathematical Reasoning [47.10036442634749]
  Curriculum learning (CL) has become a popular strategy for improving reasoning in large language models (LLMs)<n>Yet prior work employs disparate difficulty metrics and training setups, leaving open fundamental questions: When does curriculum help?<n>We address these questions through a unified offline evaluation framework that decomposes curriculum difficulty into five complementary dimensions: Problem Difficulty, Model Surprisal, Confidence Margin, Predictive Uncertainty, and Decision Variability.
  arXiv  Detail & Related papers  (2025-10-21T21:43:38Z)
• Teaching According to Talents! Instruction Tuning LLMs with Competence-Aware Curriculum Learning [64.92967672226534]
  This paper presents a Competence-Aware Multi-Perspective cUrriculum inStruction tuning framework termed CAMPUS.<n> CAMPUS offers several advantages: Dynamic selection for sub-curriculum, competency-aware adjustment to the curriculum schedule, and multiple difficulty-based scheduling.
  arXiv  Detail & Related papers  (2025-09-17T07:58:59Z)
• FAST: Similarity-based Knowledge Transfer for Efficient Policy Learning [57.4737157531239]
  Transfer Learning offers the potential to accelerate learning by transferring knowledge across tasks.<n>It faces critical challenges such as negative transfer, domain adaptation and inefficiency in selecting solid source policies.<n>In this work we challenge the key issues in TL to improve knowledge transfer, agents performance across tasks and reduce computational costs.
  arXiv  Detail & Related papers  (2025-07-27T22:21:53Z)
• Your Pretrained Model Tells the Difficulty Itself: A Self-Adaptive Curriculum Learning Paradigm for Natural Language Understanding [53.63482987410292]
  We present a self-adaptive curriculum learning paradigm that prioritizes fine-tuning examples based on difficulty scores predicted by pre-trained language models.<n>We evaluate our method on four natural language understanding (NLU) datasets covering both binary and multi-class classification tasks.
  arXiv  Detail & Related papers  (2025-07-13T19:36:17Z)
• Beyond Random Sampling: Efficient Language Model Pretraining via Curriculum Learning [23.900888224619]
  We show that curriculum learning consistently improves convergence in early and mid-training phases.<n>We identify compression ratio, lexical diversity, and readability as effective difficulty signals across settings.
  arXiv  Detail & Related papers  (2025-06-12T21:06:57Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.