English

Language Alignment via Nash-learning and Adaptive feedback

Machine Learning 2024-06-25 v1 Artificial Intelligence Computation and Language Computer Science and Game Theory

Abstract

Recent research has shown the potential of Nash Learning via Human Feedback for large language model alignment by incorporating the notion of a preference model in a minimax game setup. We take this idea further by casting the alignment as a mirror descent algorithm against the adaptive feedback of an improved opponent, thereby removing the need for learning a preference model or the existence of an annotated dataset altogether. The resulting algorithm, which we refer to as Language Alignment via Nash-learning and Adaptive feedback (LANA), is capable of self-alignment without the need for a human-annotated preference dataset. We support this statement with various experiments and mathematical discussion.

Keywords

Cite

@article{arxiv.2406.15890,
  title  = {Language Alignment via Nash-learning and Adaptive feedback},
  author = {Ari Azarafrooz and Farshid Faal},
  journal= {arXiv preprint arXiv:2406.15890},
  year   = {2024}
}

Comments

Accepted at ICML 2024 Workshop on Models of Human Feedback for AI Alignment, Vienna, Austria

R2 v1 2026-06-28T17:15:57.533Z