Papers by Haikang Deng
Reward-Augmented Decoding: Efficient Controlled Text Generation With a Unidirectional Reward Model (2023.emnlp-main)
Copied to clipboard
| Challenge: | Large language models generate text that is problematic or lacks a desired attribute. |
| Approach: | They propose a text generation procedure that uses a small unidirectional reward model to encourage a language model to generate text with certain properties. |
| Outcome: | The proposed procedure outperforms prior weighted decoding methods and matches state-of-the-art techniques that require additional training. |
Decoupling Task-Solving and Output Formatting in LLM Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Recent studies suggest intertwining task and format instructions with strict formatting requirements can negatively impact LLMs' reasoning capabilities. |
| Approach: | They propose a decoding framework that explicitly decouples format adherence from problem solving. |
| Outcome: | Experiments show that Deco-G consistently gains over prompting and structured generation baselines, with guaranteed format compliance. |