Основные понятия
SERVAL proposes a synergy learning pipeline to enhance the vertical capabilities of large language models (LLMs) and small models through mutual enhancement, achieving competitive performance in medical prediction without gold labels.
Аннотация
SERVAL introduces a novel approach to unsupervised development of vertical capabilities in LLMs and small models. By leveraging LLM's zero-shot outcomes as annotations, SERVAL enhances both models iteratively. The method shows promising results in medical diagnosis tasks, highlighting the potential of label-free training in specialized domains.
Статистика
SERVAL achieves fully-supervised competitive performance across ten medical datasets without gold labels.
Comprehensive experiments demonstrate the effectiveness of SERVAL in refining vertical capabilities.
The method progressively improves both LLMs and small models through an iterative process.
Цитаты
"SERVAL utilizes the LLM’s zero-shot outcomes as annotations, leveraging its confidence to teach a robust vertical model from scratch."
"Comprehensive experiments show that, without access to any gold labels, SERVAL attains fully-supervised competitive performance across ten widely used medical datasets."