Back to AIBriefs
AnalysisAI ModelsCybersecurity

Representation Matters in Randomized Smoothing for Audio Classification

Arxiv CS.SD (Sound/Audio ML)

This paper applies randomized smoothing to audio classification, showing that the representation space (e.g., log-mel spectrograms) critically affects certified robustness guarantees. The authors introduce a method to certify robustness despite preprocessing, achieving improved certified accuracy on several benchmarks.

·
7 days ago
Representation Matters in Randomized Smoothing for Audio Classification — AIBriefs