Timezone: »

 
Fake It Until You Make It : Towards Accurate Near-Distribution Novelty Detection
Hossein Mirzaei · Mohammadreza Salehi · Sajjad Shahabi · Efstratios Gavves · Cees Snoek · Mohammad Sabokrou · Mohammad Hossein Rohban

@
in Workshop on Machine Learning Safety »

We aim for image-based novelty detection. Despite considerable progress, existing models either fail or face dramatic drop under the so-called ``near-distribution" setup, where the differences between normal and anomalous samples are subtle. We first demonstrate existing methods could experience up to 20\% decrease in their AUCs in the near-distribution setting. Next, we propose to exploit a score-based generative model to produce synthetic near-distribution anomalous data. Our model is then fine-tuned to distinguish such data from the normal samples. We make quantitative as well as qualitative evaluation of this strategy, and compare the results with a variety of GAN-based models. Effectiveness of our method for both near-distribution and standard novelty detection is assessed through extensive experiments on datasets in diverse applications such as medical images, object classification, and quality control. This reveals that our method significantly improves upon existing models, and consistently decreases the gap between the near-distribution and standard novelty detection AUCs by a considerable amount.

[ Poster]  [ Topia] 

Author Information

Hossein Mirzaei (Sharif University of Technology)
Mohammadreza Salehi (University of Amsterdam)
Sajjad Shahabi (Sharif University of Technology)
Efstratios Gavves (University of Amsterdam)

Dr. Efstratios Gavves is an Associate Professor at the University of Amsterdam in the Netherlands, an ELLIS Scholar, and co-founder of Ellogon.AI. He is a director of the QUVA Deep Vision Lab with Qualcomm, and the POP-AART Lab with the Netherlands Cancer Institute and Elekta. Efstratios received the ERC Career Starting Grant 2020, and NWO VIDI grant 2020 to research on the Computational Learning of Time for spatiotemporal sequences and video. His background is in Computer Vision. Currently, his research interests lie in the Machine Learning of Time and Dynamics, and its applications to Vision and Sciences.

Cees Snoek (University of Amsterdam)
Mohammad Sabokrou (Institute for Research in fundamental science (IPM))
Mohammad Hossein Rohban (Sharif University of Technology)

More from the Same Authors