Paper Detail
Salman Rahman, Jingyan Shen, Anna Mordvina, Hamid Palangi, Saadia Gabriel, Pavel Izmailov
Large language models have achieved significant reasoning improvements through reinforcement learning with verifiable rewards (RLVR). Yet as model capabilities grow, constructing high-quality reward signals becomes increasingly difficult, making it essential to understand when RLVR can succeed under weaker forms of supervision. We conduct a systematic empirical study across diverse model families and reasoning domains under three weak supervision settings: scarce data, noisy rewards, and self-supervised proxy rewards. We find that generalization is governed by training reward saturation dynamics: models that generalize exhibit a prolonged pre-saturation phase during which training reward and downstream performance climb together, while models that saturate rapidly memorize rather than learn. We identify reasoning faithfulness, defined as the extent to which intermediate steps logically support the final answer, as the pre-RL property that predicts which regime a model falls into, while output diversity alone is uninformative. Motivated by these findings, we disentangle the contributions of continual pre-training and supervised fine-tuning, finding that SFT on explicit reasoning traces is necessary for generalization under weak supervision, while continual pre-training on domain data amplifies the effect. Applied together to Llama3.2-3B-Base, these interventions enable generalization across all three settings where the base model previously failed.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@article{rahman2026when,
title = {When Can LLMs Learn to Reason with Weak Supervision?},
author = {Salman Rahman and Jingyan Shen and Anna Mordvina and Hamid Palangi and Saadia Gabriel and Pavel Izmailov},
year = {2026},
abstract = {Large language models have achieved significant reasoning improvements through reinforcement learning with verifiable rewards (RLVR). Yet as model capabilities grow, constructing high-quality reward signals becomes increasingly difficult, making it essential to understand when RLVR can succeed under weaker forms of supervision. We conduct a systematic empirical study across diverse model families and reasoning domains under three weak supervision settings: scarce data, noisy rewards, and self-su},
url = {https://arxiv.org/abs/2604.18574},
keywords = {cs.LG, cs.AI, reinforcement learning with verifiable rewards, reward signals, weak supervision, reward saturation dynamics, reasoning faithfulness, supervised fine-tuning, continual pre-training, explicit reasoning traces, code available, huggingface daily},
eprint = {2604.18574},
archiveprefix = {arXiv},
}
{}