Search papers, labs, and topics across Lattice.
Aalborg University, Pioneer Centre for AI
4
0
4
You can ditch mean pooling: adaptive temporal pooling strategies for pre-trained audio embeddings dramatically improve training-free anomaly detection, even beating fully trained models.
Blindly applying anomalous sound detection models trained with machine-specific data can lead to significant performance drops when the machine identity is unknown at test time, highlighting a critical gap in current ASD benchmarks.
