Papers by Sai Selvaraj
Improving Summarization with Human Edits (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have shown the promise of learning with human feedback paradigms to produce human-determined high-quality text. |
| Approach: | They propose a novel technique to use both human-edited and model-generated data together in the training loop. |
| Outcome: | The proposed technique outperforms the conventional RLHF method (designed for human preferences) when applied to human-edit data. |