Deep Anomaly Detection Research Articles

Foundations for Unfairness in Anomaly Detection - Case Studies in Facial Imaging Data

Deep anomaly detection (AD) is perhaps the most controversial of data analytic tasks as it identifies entities that are specifically targeted for further investigation or exclusion. Also controversial is the application of AI to facial data, in particular facial recognition. This work explores the intersection of these two areas to understand two core questions: Who these algorithms are being unfair to and equally important why. Recent work has shown that deep AD can be unfair to different groups despite being unsupervised with a recent study showing that for portraits of people: men of color are far more likely to be chosen to be outliers. We study the two main categories of AD algorithms: autoencoder-based and single-class-based which effectively try to compress all the instances and those that can not be easily compressed are deemed to be outliers. We experimentally verify sources of unfairness such as the under-representation of a group (e.g people of color are relatively rare), spurious group features (e.g. men are often photographed with hats) and group labeling noise (e.g. race is subjective). We conjecture that lack of compressibility is the main foundation and the others cause it but experimental results show otherwise and we present a natural hierarchy amongst them.

Rapid identification method for on-road high-emission vehicle based on deep semi-supervised anomaly detection

This study aims to identify high-emission vehicles in urban traffic management. Existing high-emission vehicle identification models typically overlook the dispersed anomaly data distribution, and machine learning anomaly detection methods face difficulties in learning decision boundaries, leading to reduced detection performance. To address these issues, this study proposes a semi-supervised learning method based on data fusion and deep anomaly detection. This method integrates the chassis and engine dynamometer test data with on-road remote sensing system data to obtain more comprehensive vehicle emission information. Experimental results demonstrate that the proposed method introduces a penalty mechanism for anomaly samples, encouraging the model to increase the dissimilarity in similarity between normal and abnormal data at the latent data distribution level. In vehicle emission datasets from different regions, this method achieves over 95% AUC, demonstrating strong applicability and accuracy.

Weakly Supervised anomaly detection with privacy preservation under a Bi-Level Federated learning framework

Training a Machine Learning (ML) model in the industrial field faces special challenges, such as data privacy, lack of data, data imbalance, and unlabeled data. Therefore, it is not realistic to gather production data directly from various companies and use them to train a machine learning model. In this paper, we proposed a novel framework named Bi-level Federated Learning (BFL) to tackle the above challenges. In the first level, a Weakly Supervised Anomaly Detection method named Pairwise Relation prediction-based Ordinal regression Network (PRO) is utilized for training a Deep Anomaly Detection (DAD) model under Federated Learning (FL) mechanism. According to the DAD model, anomalies are identified and labeled with ‘anomaly’ tags. In order to train a multi-classification model that can identify different types of defects not just anomalies, the anomalies are manually labeled according to their respective defect types. In the second level, labeled anomaly and normal data are applied to train a multi-classification model under the FL mechanism. A framework is proposed to classify defects in pre-baked carbon anodes. Its performance is verified through a real case study. The results show that the DAD model achieved an acceptable performance with a 0.988 recall and 0.866 F1 score. The multi-classification model has shown good performance on 0.942 accuracy, 0.913 precision, 0.903 recall, and 0.906 F1. The results of the BFL exhibit improved performance compared to our previous work and classical ML methods.

Layer-Wise External Attention by Well-Localized Attention Map for Efficient Deep Anomaly Detection

The external attention mechanism offers a promising approach to enhance image anomaly detection (Hayakawa et al., in: IMPROVE, pp. 100-–110, 2023). Nevertheless, the effectiveness of this method is contingent upon the judicious selection of an intermediate layer with external attention. In this study, we performed a comprehensive series of experiments to clarify the mechanisms through which external attention improves detection performance. We assessed the performance of the LEA-Net (Hayakawa et al., in: IMPROVE, pp. 100–110, 2023), which implements layer-wise external attention, using MVTec AD and Plant Village datasets. The detection performances of the LEA-Net were compared with that of the baseline model under different anomaly maps generated by three unsupervised approaches. In addition, we investigated the relationship between the detection performance of LEA-Net and the selection of an attention point, which means an intermediate layer where external attention is applied. The findings reveal that the synergy between the dataset and the generated anomaly map influenced the effectiveness of the LEA-Net. For poorly localized anomaly maps, the selection of the attention point becomes a pivotal factor in determining detection efficiency. At shallow attention points, a well-localized attention map successfully notably improves the detection performance. For deeper attention points, the overall intensity of the attention map is essential; this intensity can be substantially amplified by layer-wise external attention, even for a low-intensity anomaly map. Overall, the results suggest that for layer-wise external attention, the positional attributes of anomalies hold greater significance than the overall intensity or visual appearance of the anomaly map.

Deep Anomaly Detection Framework Utilizing Federated Learning for Electricity Theft Zero-Day Cyberattacks.

Smart power grids suffer from electricity theft cyber-attacks, where malicious consumers compromise their smart meters (SMs) to downscale the reported electricity consumption readings. This problem costs electric utility companies worldwide considerable financial burdens and threatens power grid stability. Therefore, several machine learning (ML)-based solutions have been proposed to detect electricity theft; however, they have limitations. First, most existing works employ supervised learning that requires the availability of labeled datasets of benign and malicious electricity usage samples. Unfortunately, this approach is not practical due to the scarcity of real malicious electricity usage samples. Moreover, training a supervised detector on specific cyberattack scenarios results in a robust detector against those attacks, but it might fail to detect new attack scenarios. Second, although a few works investigated anomaly detectors for electricity theft, none of the existing works addressed consumers' privacy. To address these limitations, in this paper, we propose a comprehensive federated learning (FL)-based deep anomaly detection framework tailored for practical, reliable, and privacy-preserving energy theft detection. In our proposed framework, consumers train local deep autoencoder-based detectors on their private electricity usage data and only share their trained detectors' parameters with an EUC aggregation server to iteratively build a global anomaly detector. Our extensive experimental results not only demonstrate the superior performance of our anomaly detector compared to the supervised detectors but also the capability of our proposed FL-based anomaly detector to accurately detect zero-day attacks of electricity theft while preserving consumers' privacy.

Deep-Shallow Metaclassifier with Synthetic Minority Oversampling for Anomaly Detection in a Time Series

Anomaly detection in data streams (and particularly time series) is today a vitally important task. Machine learning algorithms are a common design for achieving this goal. In particular, deep learning has, in the last decade, proven to be substantially more accurate than shallow learning in a wide variety of machine learning problems, and deep anomaly detection is very effective for point anomalies. However, deep semi-supervised contextual anomaly detection (in which anomalies within a time series are rare and none at all occur in the algorithm’s training data) is a more difficult problem. Hybrid anomaly detectors (a “normal model” followed by a comparator) are one approach to these problems, but the separate loss functions for the two components can lead to inferior performance. We investigate a novel synthetic-example oversampling technique to harmonize the two components of a hybrid system, thus improving the anomaly detector’s performance. We evaluate our algorithm on two distinct problems: identifying pipeline leaks and patient-ventilator asynchrony.

Unmasking overestimation: a re-evaluation of deep anomaly detection in spacecraft telemetry

As the volume of telemetry data generated by satellites and other complex systems continues to grow, there is a pressing need for more efficient and accurate anomaly detection methods. Current techniques often rely on human analysis and preset criteria, presenting several challenges including the necessity for expert interpretation and continual updates to match the dynamic mission environment. This paper critically examines the use of deep anomaly detection (DAD) methods in addressing these challenges, evaluating their efficacy on real-world spacecraft telemetry data. It exposes limitations in current DAD research, highlighting the tendency for performance results to be overestimated and suggesting that simpler methods can sometimes outperform more complex DAD algorithms. By comparing established metrics for anomaly detection with newly proposed ones, this paper aims to improve the evaluation of DAD algorithms. It underscores the importance of using less accuracy-inflating metrics and offers a comprehensive comparison of DAD methods on popular benchmark datasets and real-life satellite telemetry data. Among the DAD methods examined, the LSTM algorithm demonstrates considerable promise. However, the paper also reveals the potential limitations of this approach, particularly in complex systems that lack a single, clear predictive failure channel. The paper concludes with a series of recommendations for future research, including the adoption of best practices, the need for high-quality, pre-split datasets, and the investigation of other prediction error methods. Through these insights, this paper contributes to the improved understanding and application of DAD methods, ultimately enhancing the reliability and effectiveness of anomaly detection in real-world scenarios.

Unsupervised Deep Anomaly Detection for Industrial Multivariate Time Series Data

With the rapid development of deep learning, researchers are actively exploring its applications in the field of industrial anomaly detection. Deep learning methods differ significantly from traditional mathematical modeling approaches, eliminating the need for intricate mathematical derivations and offering greater flexibility. Deep learning technologies have demonstrated outstanding performance in anomaly detection problems and gained widespread recognition. However, when dealing with multivariate data anomaly detection problems, deep learning faces challenges such as large-scale data annotation and handling relationships between complex data variables. To address these challenges, this study proposes an innovative and lightweight deep learning model—the Attention-Based Deep Convolutional Autoencoding Prediction Network (AT-DCAEP). The model consists of a characterization network based on convolutional autoencoders and a prediction network based on attention mechanisms. The AT-DCAEP exhibits excellent performance in multivariate time series data anomaly detection without the need for pre-labeling large-scale datasets, making it an efficient unsupervised anomaly detection method. We extensively tested the performance of AT-DCAEP on six publicly available datasets, and the results show that compared to current state-of-the-art methods, AT-DCAEP demonstrates superior performance, achieving the optimal balance between anomaly detection performance and computational cost.

Deep Anomaly Detection with Attention (DADA): A Novel Approach for Identifying Multipath Interference in Radar Signals

Multipath interference in radar signals caused by sea, ground, and other environments poses significant challenges to the target detection, tracking, and classification capabilities of radar systems. Existing methods for radar signal identification require labeled samples and focus mainly on the classification of normal signals. However, in practice, anomalous samples (multipath interference signals) may be scarce and highly imbalanced (i.e., mostly normal samples). To address this problem, we propose a deep anomaly detection with attention (DADA) for semisupervised detection of multipath radar signals. The method transforms radar signals into time–frequency images and is trained exclusively on normal samples. The autoencoder architecture is extended with a feature extractor network to capture latent sample features. CBAM attention is introduced to improve feature extraction. By learning the distribution of normal samples in high-dimensional image space and low-dimensional feature space, a two-dimensional feature space representing normal samples is constructed. A one-class SVM then learns the boundary of normal samples for anomaly detection. Extensive experiments on radar signal datasets validate the effectiveness of the proposed approach.

Deep Anomaly Detection of Voltage Waveform Distortion in the Transmission Grid Due to Geomagnetically Induced Currents

Deep Anomaly Detection of Voltage Waveform Distortion in the Transmission Grid Due to Geomagnetically Induced Currents

A Deep Anomaly Detection System for IoT-Based Smart Buildings

In recent years, technological advancements in sensor, communication, and data storage technologies have led to the increasingly widespread use of smart devices in different types of buildings, such as residential homes, offices, and industrial installations. The main benefit of using these devices is the possibility of enhancing different crucial aspects of life within these buildings, including energy efficiency, safety, health, and occupant comfort. In particular, the fast progress in the field of the Internet of Things has yielded exponential growth in the number of connected smart devices and, consequently, increased the volume of data generated and exchanged. However, traditional Cloud-Computing platforms have exhibited limitations in their capacity to handle and process the continuous data exchange, leading to the rise of new computing paradigms, such as Edge Computing and Fog Computing. In this new complex scenario, advanced Artificial Intelligence and Machine Learning can play a key role in analyzing the generated data and predicting unexpected or anomalous events, allowing for quickly setting up effective responses against these unexpected events. To the best of our knowledge, current literature lacks Deep-Learning-based approaches specifically devised for guaranteeing safety in IoT-Based Smart Buildings. For this reason, we adopt an unsupervised neural architecture for detecting anomalies, such as faults, fires, theft attempts, and more, in such contexts. In more detail, in our proposal, data from a sensor network are processed by a Sparse U-Net neural model. The proposed approach is lightweight, making it suitable for deployment on the edge nodes of the network, and it does not require a pre-labeled training dataset. Experimental results conducted on a real-world case study demonstrate the effectiveness of the developed solution.

A Deep Anomaly Detection With Same Probability Distribution and Its Application in Rolling Bearing

Abstract An innovative deep-learning-based model, namely, deep anomaly detection with the same probability distribution (DADSPD) is proposed to improve the accuracy of anomaly detection (AD) of rolling bearings driven only by normal data. First, the main framework of feature extraction based on a residual network was established, and a three-layer encoder structure was used to extract multidimensional features. Second, a new loss function based on the same probability distribution is designed, and the function of its probability distribution is to complete the training of the model by calculating the similarity between the outputs. Subsequently, the vibration data were preprocessed using wavelet and envelope analysis, and the processed data are converted into two-dimensional image signals and used as the input of the DADSPD. Finally, the model is verified on three sets of run-to-failure experimental datasets of rolling bearing. The results demonstrate that the proposed DADSPD model reaches more than 99%, which indicates that the DADSPD model has a high fault early warning and AD capability.

Deep Anomaly Detection for Automotive Components by Oversampling

Deep Anomaly Detection for Automotive Components by Oversampling

Deep Anomaly Detection and Search via Reinforcement Learning (Student Abstract)

Deep Anomaly Detection and Search via Reinforcement Learning (Student Abstract)

A Quantum-Classical Hybrid Solution for Deep Anomaly Detection.

Machine learning (ML) has achieved remarkable success in a wide range of applications. In recent ML research, deep anomaly detection (AD) has been a hot topic with the aim of discriminating among anomalous data with deep neural networks (DNNs). Notably, image AD is one of the most representative tasks in current deep AD research. ML's interaction with quantum computing is giving rise to a heated topic named quantum machine learning (QML), which enjoys great prospects according to recent academic research. This paper attempts to address the image AD problem in a deep manner with a novel QML solution. Specifically, we design a quantum-classical hybrid DNN (QHDNN) that aims to learn directly from normal raw images to train a normality model and then exclude images that do not conform to this model as anomalies during its inference. To enable the QHDNN to perform satisfactorily in deep image AD, we explore multiple quantum layer architectures and design a VQC-based QHDNN solution. Extensive experiments were conducted on commonly used benchmarks to test the proposed QML solution, whose results demonstrate the feasibility of addressing deep image AD with QML. Importantly, the experimental results show that our quantum-classical hybrid solution can even yield superior performance to that of its classical counterpart when they share the same number of learnable parameters.

An innovative deep anomaly detection of building energy consumption using energy time-series images

Deep anomaly detection (DAD) is essential in optimizing building energy management. Nonetheless, most existing works concerning this field consider unsupervised learning and involve the analysis of sensor readings through a one-dimensional (1D) energy time series, which limits the options for detecting anomalies within the building’s energy consumption. To the best of the authors’ knowledge, this paper presents the first study that explores using two-dimensional (2D) image representations as features of a supervised deep transfer learning (DTL) approach. Specifically, using 2D image representations allows taking advantage of any underlying data within the feature set, providing more possibilities to encode data and detect pertinent features which may not be considered in standard 1D time-series. Furthermore, the effects of using CNN activations as machine learning (ML) model features are also investigated to combine the advantages of both techniques. Additionally, the concept of layer and hyperparameter variation for the CNN model is also studied, with the objective of reducing the overall time computation and resource requirements of the proposed system. Hence, this makes our approach a candidate for edge-based applications. As per the conducted experiments, the top methodology rests at the F1-scores of 93.63% and 99.89% under simulated and real-world energy datasets, respectively. This involves using grayscale 2D image representations that combine CNN activations extracted from AlexNet and GoogleNet pre-trained models as features to a linear support vector machine (SVM) classifier. Finally, the comparison analysis with the state-of-the-art has shown the superiority of the proposed method in various assessment criteria.

Real-Time Deep Anomaly Detection Framework for Multivariate Time-Series Data in Industrial IoT

The data produced by millions of connected devices and smart sensors in the Industrial Internet of Things (IIoT) is highly dynamic, large-scale, heterogeneous, and time-stamped. These time-stamped data are the core of IIoT automation and have the potential to affect industrial processes intensely. It poses significant challenges to effectively detect anomalies from time-series data and deliver actionable insights in real time to drive improvements to industrial processes. In most practical applications, where data are used to make automated decisions, real-time anomaly detection is critical. With this focus, in this article, we advise a hybrid end-to-end deep anomaly detection (DAD) framework to accurately detect anomalies and extremely rare events on sensitive, Internet of Things (IoT) streaming data in real time or near real time. The proposed framework is based on a convolutional neural network (CNN) and a two-stage long short-term memory (LSTM)-based Autoencoder (AE). We exploit a two-stage LSTM AE in parallel to detect anomalies and extremely rare events hidden in massive sensor data by identifying short- and long-term variations in actual sensor values from the predicted values. We design and train a hybrid model using the Keras/TensorFlow framework as the backend. The experimental results on one simulation and two real datasets demonstrate that the proposed framework achieved better performance and outperforms other state-of-the-art competitive models. Moreover, to prove that the proposed model can be designed for the network edge, we train, optimize, and quantize the model to run-on resource-constrained (i.e., edge) devices. Further evaluation indicates that the training and inference time for each sample is short enough to carry out anomaly detection on edge.

Deep Anomaly Detection for In-Vehicle Monitoring—An Application-Oriented Review

Anomaly detection has been an active research area for decades, with high application potential. Recent work has explored deep learning approaches to the detection of abnormal behaviour and abandoned objects in outdoor video surveillance scenarios. The extension of this recent work to in-vehicle monitoring using solely visual data represents a relevant research opportunity that has been overlooked in the accessible literature. With the increasing importance of public and shared transportation for urban mobility, it becomes imperative to provide autonomous intelligent systems capable of detecting abnormal behaviour that threatens passenger safety. To investigate the applicability of current works to this scenario, a recapitulation of relevant state-of-the-art techniques and resources is presented, including available datasets for their training and benchmarking. The lack of public datasets dedicated to in-vehicle monitoring is addressed alongside other issues not considered in previous works, such as moving backgrounds and frequent illumination changes. Despite its relevance, similar surveys and reviews have disregarded this scenario and its specificities. This work initiates an important discussion on application-oriented issues, proposing solutions to be followed in future works, particularly synthetic data augmentation to achieve representative instances with the low amount of available sequences.

Inline nondestructive internal disorder detection in pear fruit using explainable deep anomaly detection on X-ray images

To preserve the quality of fresh fruit after harvest and to meet the year-round demand for high-quality fruit, pears are stored under a controlled atmosphere. However, due to preharvest events or suboptimal storage conditions, internal disorders might develop resulting in severe quality loss. Examples include internal browning and cavities, which are invisible externally. Here, X-ray radiography is investigated as a technique for internal quality inspection. The detection of defect fruit is approached as an anomaly detection (AD) problem, in which a model is constructed using nominal data and an anomaly score is used to identify defect fruit. In this work, multiple deep AD methods are shown to be effective to detect pears with internal cavity and browning disorders using X-ray radiographs (mean area under the receiver operating characteristic curve (AUC) up to 0.962). The best performing methods were found on par with a state-of-the-art multisensor disorder detection method (mean AUC up to 0.966). By investigating AD performance in function of internal disorder severity, it was shown that defect fruit with a cavity volume percentage > 1.0% could be detected 100% accurate using inline X-ray imaging. For lower cavity area percentages, the accuracy depended on the internal browning severity. Additionally, the explainability of the deep AD methods, i.e., how well human interpretable insight can be provided from each method’s predictions, were qualitatively evaluated using anomaly heatmaps, which provided useful insight in the execution of the deep learning algorithms.

Unsupervised Deep Anomaly Detection for Medical Images Using an Improved Adversarial Autoencoder.

Anomaly detection has been applied in the various disease of medical practice, such as breast cancer, retinal, lung lesion, and skin disease. However, in real-world anomaly detection, there exist a large number of healthy samples, and but very few sick samples. To alleviate the problem of data imbalance in anomaly detection, this paper proposes an unsupervised learning method for deep anomaly detection based on an improved adversarial autoencoder, in which a module called chain of convolutional block (CCB) is employed instead of the conventional skip-connections used in adversarial autoencoder. Such CCB connections provide considerable advantages via direct connections, not only preserving both global and local information but also alleviating the problem of semantic disparity between the encoding features and the corresponding decoding features. The proposed method is thus able to capture the distribution of normal samples within both image space and latent vector space. By means of minimizing the reconstruction error within both spaces during training phase, higher reconstruction error during test phase is indicative of an anomaly. Our method is trained only on the healthy persons in order to learn the distribution of normal samples and can detect sick samples based on high deviation from the distribution of normality in an unsupervised way. Experimental results for multiple datasets from different fields demonstrate that the proposed method yields superior performance to state-of-the-art methods.