Can I read 1456 Sample Efficient Preferen on EtoBox?
1456 Sample Efficient Preferen by jill.hsujing is a document available to read on EtoBox.
What is 1456 Sample Efficient Preferen about?
The paper presents an active exploration algorithm for preference alignment in large language models (LLMs), formalizing the problem as a dueling contextual bandit scenario. It introduces efficient methods for selecting human feedback data, demonstrating improved performance over baseline methods with limited samples across multiple datasets. The authors also contribute two new datasets and theoretical guarantees for their approach, which enhances sample efficiency in preference-based learning tasks.
- Author
- jill.hsujing
- Language
- EN