RanAT4BIE: Random Adversarial Training for Biomedical Information Extraction

Abstract--We introduce random adversarial training (RA T), a novel framework successfully applied to biomedical information extraction (BioIE) tasks. While adversarial training yields significant improvements across various performance metrics, it also introduces considerable computational overhead. T o address this limitation, we propose RA T as an efficiency solution for biomedical information extraction. Through comprehensive evaluations, RA T demonstrates superior performance compared to baseline models in BioIE tasks. Adversarial training was initially conceptualized as a methodology for enhancing the robustness of deep learning models [1].