Detecting semantic anomalies

By Faruk Ahmed, Aaron C. Courville. 2019

98 citations recorded by the source corpus.

We critically appraise the recent interest in out-of-distribution (OOD) detection and question the practical relevance of existing benchmarks. While the currently prevalent trend is to consider different datasets as OOD, we argue that out-distributions of practical interest are ones where the distinction is semantic in nature for a specified context, and that evaluative tasks should reflect this more closely. Assuming a context of object recognition, we recommend a set of benchmarks, motivated by practical applications. We make progress on these benchmarks by exploring a multi-task learning based approach, showing that auxiliary objectives for improved semantic awareness result in improved semantic anomaly detection, with accompanying generalization benefits.

Subfields: Artificial Intelligence, Anomaly & Outlier Detection

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Anomaly & Outlier Detection

Citations in the corpus

CODEs: Chamfer Out-of-Distribution Examples against Overconfidence Issue (2021, background): “We evaluate the situation where in-distribution samples are not only significantly outnumber OOD ones, but also have significant semantic shifts following [1].”
Bayesian Variational Autoencoders for Unsupervised Out-of-Distribution Detection (2019, unknown): “…et al., 2018; Ren et al., 2019; Zenati et al., 2018; Akcay et al., 2018), and (b) eight classes of FashionMNIST (in-distribution) vs. the remaining two classes (OoD), using ﬁve different splits { (0 , 1) , (2 , 3) , (4 , 5) , (6 , 7) , (8 , 9) } of held-out classes (Ahmed & Courville, 2019).”
A Unified Model for Multi-class Anomaly Detection (2022, background): “Semantic AD [1, 4] also involves multiple classes, which mainly focus on anomaly detection extended from classification datasets ( e.g. , CIFAR-10 CIFAR-10 First, we focus more on the industrial anomaly detection dataset, MVTec-AD [2], which is of more practical usage.”
Generalized Out-of-Distribution Detection: A Survey (2021, unknown): “…Oursensory/semantic dichotomy for the AD sub-task deﬁnition comes from the low-level sensory anomalies and high-level semantic anomalies that are introduced in Ahmed and Courville (2020) and highlighted in the recent AD survey (Ruff et al., 2021), to reﬂect the rise of deep learning.”
OpenOOD: Benchmarking Generalized Out-of-Distribution Detection (2022, background): “According to different distribution shifts that causes the anomalies, AD tasks can be further divided into sensory AD that to detect low-level sensory anomalies, and semantic AD that to detect high-level semantic anomalies [36, 37].”
A Unifying Review of Deep and Shallow Anomaly Detection (2020, methodology): “Although these synthetic protocols have some value, it has been questioned how well they reflect real progress on challenging AD tasks [200], [332].”