Revisiting Chain-of-Thought Reasoning under Limited Supervision: Semi-supervised Chain-of-Thought Learning
Leveraging unlabeled data to refine model reasoning through entropy-gated pseudo-supervision.
Chain-of-thought (CoT) reasoning has emerged as an effective approach for activating latent reasoning capabilities in large language models. However, most existing CoT methods use reasoning chains mainly as inference-time prompts, while the generated reasoning traces are rarely r...
ArXiv CS.AI
Source
Generated insight by OnAirToday AI
This research shifts the paradigm of chain-of-thought reasoning by demonstrating that unlabeled data can be repurposed into high-precision pseudo-supervision signals. By filtering reasoning chains through semantic entropy, the framework offers a scalable path to improving model reasoning without the high cost of human-labeled demonstrations.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.