An end-to-end convolutional autoencoder plus a convolutional alarm network, trained jointly on reconstruction and classification losses, achieves near-state-of-the-art anomaly detection AUC on MNIST, Fashion-MNIST, CIFAR-10, and several tabular datasets.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
End-to-End Convolutional Activation Anomaly Analysis for Anomaly Detection
An end-to-end convolutional autoencoder plus a convolutional alarm network, trained jointly on reconstruction and classification losses, achieves near-state-of-the-art anomaly detection AUC on MNIST, Fashion-MNIST, CIFAR-10, and several tabular datasets.