Anomaly Detection With Multiple-Hypotheses Predictions

Authors: Duc Tam Nguyen, Zhongyu Lou, Michael Klar, Thomas Brox

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the anomaly detection performance of our approach on CIFAR-10 and a real anomaly image dataset, the Metal Anomaly dataset... We show that anomaly detection performance with multiple-hypotheses networks is significantly better compared to single-hypotheses networks. On CIFAR-10, our proposed Con AD framework (consistency-based anomaly detection) improves on previously published results.
Researcher Affiliation Collaboration 1Computer Vision Group, University of Freiburg, Freiburg, Germany 2Corporate Research, Robert Bosch Gmb H, Renningen, Germany.
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository.
Open Datasets Yes We evaluate the anomaly detection performance of our approach on CIFAR-10 and a real anomaly image dataset, the Metal Anomaly dataset
Dataset Splits Yes Table 1. Dataset description. ... NORMAL DATA TRAIN 4500 5408 VALID 500 1352 TEST 1000 1324
Hardware Specification No The paper does not provide specific details about the hardware used, such as GPU or CPU models.
Software Dependencies No The paper mentions software like 'Adam' and 'DCGAN' but does not specify version numbers for any key software components or libraries.
Experiment Setup Yes The batch-size n was set to 64 each on CIFAR-10, 32 on the Metal Anomaly dataset. Adam (Kingma & Ba, 2014) was used for training with a learning rate of 0.001. Per discriminator training, the generator is trained at most five epochs to balance both players.