AI Sec Watch: A Security Intelligence Platform for AI Systems

Luu, T.J.

Enhanced Masking-Differential Prompting (MDP): Defending Backdoor Attacks for Pre-trained Language Models Under Few-Shot Learning

inforesearchPeer-ReviewedLLM-Specific

securityresearch

Source: IEEE Xplore (Security & AI Journals)December 22, 2025

Summary

Pre-trained language models (PLMs, large AI systems trained on text data) can be vulnerable to backdoor attacks, where hidden triggers in input cause the model to produce manipulated output. This paper proposes an enhanced defense method called masking-differential prompting (MDP) that works with few-shot learning (training on very small datasets), using Jensen-Shannon divergence (a mathematical measure to compare probability distributions) instead of traditional methods and an automatic threshold-selection approach to better detect and block these attacks.

Solution / Mitigation

The paper proposes two enhancements to the masking-differential prompting (MDP) defense method: (1) adopting Jensen–Shannon (JS) divergence instead of Kullback–Leibler (KL) divergence to handle cases where anchor set information has insufficient density, keeping the divergence finite and better exploiting available data; and (2) proposing an adaptive threshold method that automatically searches for the threshold based on false rejection rate (FRR) allowance, replacing the computationally expensive manual threshold selection method using ROC curve (AUC).