Joan Serrà,David Álvarez,Vicenç Gómez,Olga Slizovskaia,José F. Núñez,Jordi Luque
ICLR, (2020)
We show that input complexity has a strong effect in those likelihoods, and pose that it is the main culprit for the puzzling results of using generative models’ likelihoods for OOD detection
This paper presents an adversarially learned approach in which both the generator and the discriminator are utilized to perform a stable and robust anomaly detection
Our further analysis using a larger scale image dataset shows that the data with only semantic shift is harder to detect, pointing out a challenge for future works to address
In this paper, inspired by the fact that differential privacy implies stability, we apply DP noise to improve the performance of outlier detection and novelty detection, with an extension to backdoor attack detection
To provide insights into prior results, part of our discussion has focused on an in-depth exploration of the popular class of normalizing flows based on affine coupling layers
We have shown that framing video anomaly detection as a self-training deep ordinal regression task overcomes some of the key limitations of existing approaches to this important problem
To optimize the objective function under the unsupervised setting, we investigate the condition to bypass the third term and get a lower bound on the objective function which can be considered as a trade-off between the mutual information and the entropy
A normalcy suppression mechanism is proposed which collaborates with the backbone network in detecting anomalies by learning to suppress the features corresponding to the normal portions of an input video
Our result is to be compared with the recent discovery that energy-based models assign lower likelihood to outliers under this setting, which naturally leads to the question of whether a calibrated deep generative models should always have a similar behavior
We proposed OOD-model-agnostic meta learning, which is a meta-learning method used for implementing K-shot N -way classification and OOD detection simultaneously