Conference Paper Published

Few-Shot Learning for Syndrome Differentiation with Two Prompts

Zhou, X., Lee, S. Y. M., Gang, X., Yang, Y., Liu, Y., Ju, S., & Li, S.* (2025). Few-Shot Learning for Syndrome Differentiation with Two Prompts. In W. Senn, M. Sanguineti, A. Saudargiene, I. V. Tetko, A. E. P. Villa, V. Jirsa, & Y. Bengio (Eds.), Artificial Neural Networks and Machine Learning – ICANN 2025: 34th International Conference on Artificial Neural Networks, Kaunas, Lithuania, September 9–12, 2025, Proceedings, Part IV, 39-50. (Lecture Notes in Computer Science; Vol. 16071 LNCS).

DOI: https://doi.org/10.1007/978-3-032-04555-3_4

Abstract

Currently, the majority of approaches to syndrome differentiation in Traditional Chinese Medicine (TCM) conceptualize this endeavor as a text classification challenge. The state-of-the-art technique in text classification is the few-shot fine-tuning approach grounded in large language models. This approach minimizes the need for extensive manual annotation while yielding remarkable classification outcomes with a limited number of labeled samples. In this paper, we propose a few-shot fine-tuning approach that integrates two distinct types of prompts tailored for the syndrome differentiation task. One type of prompt directly inquires about the syndrome type, while the other indirectly seeks to infer the syndrome type. Specifically, our approach initially adopts the prefix-tuning approach, which utilizes two prefix vectors to combine these two types of prompts. Subsequently, we fine-tune the prefix-related parameters and combine the results from the two prefix fine-tunings to generate the final output. Experimental results demonstrate the superiority of our proposed approach.

Keywords
Few-shot learning,Syndrome differentiation,Traditional Chinese Medicine