Applications of Bayesian Neural Networks in Outlier Detection

Chen, Tao

doi:10.1089/big.2021.0343

Cited by 3 publications

(3 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, ALOE (Chen et al 2022) trained the model with adversarial outliers generated within the ϵ-ball. Other methods (Mohseni et al 2020;Hendrycks et al 2020Khalid et al 2022;Ming, Fan, and Li 2022;Tao et al 2023;Du et al 2022) proposed to regularize image classifiers via self-supervised training. However, synthetic outliers may be sampled from imprecise decision boundaries.…”

Section: Related Workmentioning

confidence: 99%

Quantile-Based Maximum Likelihood Training for Outlier Detection

Taghikhah,

Kumar,

Šegvić

et al. 2024

AAAI

View full text Add to dashboard Cite

Discriminative learning effectively predicts true object class for image classification. However, it often results in false positives for outliers, posing critical concerns in applications like autonomous driving and video surveillance systems. Previous attempts to address this challenge involved training image classifiers through contrastive learning using actual outlier data or synthesizing outliers for self-supervised learning. Furthermore, unsupervised generative modeling of inliers in pixel space has shown limited success for outlier detection. In this work, we introduce a quantile-based maximum likelihood objective for learning the inlier distribution to improve the outlier separation during inference. Our approach fits a normalizing flow to pre-trained discriminative features and detects the outliers according to the evaluated log-likelihood. The experimental evaluation demonstrates the effectiveness of our method as it surpasses the performance of the state-of-the-art unsupervised methods for outlier detection. The results are also competitive compared with a recent self-supervised approach for outlier detection. Our work allows to reduce dependency on well-sampled negative training data, which is especially important for domains like medical diagnostics or remote sensing.

show abstract

Section: Related Workmentioning

confidence: 99%

Quantile-Based Maximum Likelihood Training for Outlier Detection

Taghikhah,

Kumar,

Šegvić

et al. 2024

AAAI

View full text Add to dashboard Cite

show abstract

“…Instead of treating each data stream independently. Neural network, with their different types and developments, has been used in a wide variety of research into anomaly detection [29] [30] [31] [32]]. FastABOD [22] Fast Angle-Based Outlier Detection using approximation COPOD [23] COPOD: Copula-Based Outlier Detection…”

Section: B4 Neural Networkmentioning

confidence: 99%

TTDD: a time series Tester, Transformer, and Decomposer framework for outlier Detection

Khattab

2023

Journal of Engineering Research

View full text Add to dashboard Cite

Anomaly detection in time series has become an important aspect of data analysis and has many applications. Anomaly detection is often a challenge for statistical and pattern detection modeling. We introduce TTDD (Test, Transform, Decompose, and Detection), a pattern-based detector to detect outlier values in time series datasets. TTDD splits each time series into three components, each representing an underlying pattern category. Trend, seasonality, and residual. The outlier is determined for each component separately, which contributes to determining the outlier with high accuracy. The lag in the timing between the test set and the training set was discovered using Fourier transforms. In contrast to previous work, we support time series that are phase shifted. TTDD is resilient to consecutively outlier values with different types. The experimental results are based on real-world and benchmark datasets. The proposed TTDD framework outperforms the other methods with higher accuracy score at 95% which is higher than those of other methods about 15%. Furthermore, it can detect the delay time outliers that other methods cannot easily identify.

show abstract

“…For CI-FAR(Krizhevsky and Hinton 2009) benchmarks, CIFAR10 and CIFAR100 were respectively used as the ID datasets, and six datasets were used as OOD test sets, including Textures(Cimpoi et al 2014), SVHN(Netzer et al 2011), iSUN(Xu et al 2015), Places365(Zhou et al 2017), LSUN-C(Yu et al 2015), and LSUN-R(Yu et al 2015). For largescale ImageNet benchmarks, two different sets of 100 Im-ageNet(Deng et al 2009) classes, namely ImageNet100-I and ImageNet100-II(Tao et al 2023), were used as ID sets considering that both sets have been used in related literature, and four OOD test datasets, Places(Zhou et al 2017), Textures, iNaturalist (Van Horn et al 2018, and SUN(Xiao et al 2010) were used for evaluation. There are no overlapped classes between OOD datasets and corresponding ID datasets.…”

mentioning

confidence: 99%

TagFog: Textual Anchor Guidance and Fake Outlier Generation for Visual Out-of-Distribution Detection

Chen,

Zhang,

Zheng

et al. 2024

AAAI

View full text Add to dashboard Cite

Out-of-distribution (OOD) detection is crucial in many real-world applications. However, intelligent models are often trained solely on in-distribution (ID) data, leading to overconfidence when misclassifying OOD data as ID classes. In this study, we propose a new learning framework which leverage simple Jigsaw-based fake OOD data and rich semantic embeddings (`anchors') from the ChatGPT description of ID knowledge to help guide the training of the image encoder. The learning framework can be flexibly combined with existing post-hoc approaches to OOD detection, and extensive empirical evaluations on multiple OOD detection benchmarks demonstrate that rich textual representation of ID knowledge and fake OOD knowledge can well help train a visual encoder for OOD detection. With the learning framework, new state-of-the-art performance was achieved on all the benchmarks. The code is available at https://github.com/Cverchen/TagFog.

show abstract

Applications of Bayesian Neural Networks in Outlier Detection

Cited by 3 publications

References 12 publications

Quantile-Based Maximum Likelihood Training for Outlier Detection

Quantile-Based Maximum Likelihood Training for Outlier Detection

TTDD: a time series Tester, Transformer, and Decomposer framework for outlier Detection

TagFog: Textual Anchor Guidance and Fake Outlier Generation for Visual Out-of-Distribution Detection

Contact Info

Product

Resources

About