New method boosts ASR model defense against adversarial attacks

Researchers have developed a new method called Precision-Varying Prediction (PVP) to enhance the adversarial robustness of automatic speech recognition (ASR) systems. By randomly altering the precision of an ASR model during inference, PVP significantly reduces the success rate of adversarial attacks. This technique can also be used to detect adversarial examples by classifying differences in model outputs run at varying precisions. When combined with existing uncertainty-based defenses, PVP demonstrates improved security against adaptive threats across various ASR models, languages, and attack types. AI

IMPACT Enhances the security and reliability of speech recognition systems against malicious manipulation.

RANK_REASON The cluster contains an academic paper detailing a new research method. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Hugging Face
• Matías Pizarro
• Precision-Varying Prediction (PVP)

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →