Welcome to click the blue word above, pay attention~
Related Reading:
[1] How about the generation of GAN on medical images?
Virtual change of clothes! Take a quick look at what these latest papers are doing! \
Face makeup migration! A quick look at several papers made with GAN \
GAN has been around for 6 years! It's time for a stroke! \
01-The Concise Principle of GAN Formula - Iron Armored Little Treasures
Record today, some papers using GAN for anomaly detection ! \
Anomaly detection, a very common problem.
In terms of images, for example, when you enter and exit the subway security check every day, you often see little sisters and brothers sitting there staring at your luggage through the inspection image, similar to the following (the picture comes from the GANomaly paper):
For example, in some medical image analysis, images from healthy people may be easier to obtain, and the "pattern" of the images is often fixed or not changing, while the number of images of lesions is small and difficult to obtain, or When the lesion area is changeable or even unknown, anomaly detection faces the situation that there are few positive samples/abnormal images, and relatively, normal images are easier to obtain. This situation is actually reflected in many scenarios, such as industrial visual inspection and so on.
For a large number of known categories, whether abnormal or not, we may be able to solve it by training a classification model. But in the face of perhaps unknown and variable situations, it seems difficult to tell them apart with a multi-class model. If you just want to distinguish whether it is abnormal, maybe you can do a single classifier.
We try our best to let the model fully learn what the distribution of normal data looks like. Once an abnormal image comes, even if it does not know what the new distribution is, it can still confidently tell you: I have never seen this thing before, this Heterogeneous!
How to do some network with GAN? The general idea is:
In the case of only negative samples (normal data) or a small number of positive samples:
Training phase:
Only the data distribution of negative samples (normal data) can be learned through the network, and the resulting model G can only generate or reconstruct normal data.
Test phase:
使用测试样本输入训练好的模型G，如果G经过重建后输出和输入一样或者接近，表明测试的是正常数据，否则是异常数据。
模型G的选择：
一个重建能力或者学习数据分布能力较好的生成模型，例如GAN或者VAE，甚至encoder-decoder。
下面速览几篇论文、看看GAN是如何做异常检测的（数据主要为图像形式）：
1. IPMI 2017 AnoGAN ( Unsupervised Anomaly Detection with Generative Adversarial Networks to Guide Marker Discovery )
思路：通过一个GAN的生成器G来学习正常数据的分布，测试时图像通过学习到的G找到它应该的正常图的样子，再通过对比来找到异常与否的情况。
如上图所示，AnoGAN论文中采用的是DCGAN，一种较简单的GAN架构。
训练阶段：
对抗训练，从一个噪声向量Z通过几层反卷积搭建的生成器G学习生成正常数据图像。
测试阶段：
随机采样一个高斯噪声向量z，想要通过已经训练好的G生成一幅和测试图像x对应的正常图像G(z)。G的参数是固定的，它只能生成落在正常数据分布的图像。但此时仍需进行训练，把z看成待更新的参数，通过比较G(z)和x的差异去更新，从而生成一个与x尽可能相似、理想对应的正常图像。
如果x是正常的图像，那么x和G(z)应该是一样的。
如果x异常，通过更新z，可以认为重建出了异常区域的理想的正常情况，这样两图一对比不仅仅可以认定异常情况，同时还可以找到异常区域。
为了比较G(z)和x差异去更新z：
一是通过计算G(z)和x的图像层面的L1 loss：
二是利用到训练好的判别器D，取G(z)和x在判别器D的中间层的特征层面的loss：
两者综合：
另外，异常分数计算方法：
2. 2018-02 EFFICIENT GAN-BASED ANOMALY DETECTION
针对AnoGAN测试阶段仍然需要更新参数的缺陷，此方法提出一种基于BiGAN可快百倍的方法。
训练时，同时学习将输入样本x映射到潜在表示z的编码器E，以及生成器G和判别器D:
如此可避免测试仍需要“找到z”那个耗时的步骤。与常规GAN中的D仅考虑输入（实际的或生成的）图像不同，而还考虑了潜在表示z(作为输入)。
测试时，判断图像的异常与否的分值计算方法，可选择可AnoGAN基本一样的方法。
3. 2018-12 Adversarially Learned Anomaly Detection
第二种方法的加强版，也是基于BiGAN,并且在稳定训练上做了些功夫。如下所示，（乖乖，搞了三个判别器 =_=
检测时的计算方法：
4. 2018-11-13 GANomaly: Semi-Supervised Anomaly Detection via Adversarial Training
原理：
训练时，约束正常的数据编码得到潜在空间表示z1，和对z1解码、再编码得到的z2，差距不会特别大，理想应该是一样的。
所以训练好后，用正常样本训练好的 G只能重建正常数据分布，一旦用于从未见过的异常样本编码、解码、再经历编码得到的潜在空间Z差距是大的。
当两次编码得到的潜在空间差距大于一定阈值的时候，就判定样本是异常样本。
5. 2019-01-25 Skip-GANomaly: Skip Connected and Adversarially Trained Encoder-Decoder Anomaly Detection
6. PRICAI 2018 A Surface Defect Detection Method Based on Positive Samples
原理：
C（x~|x）是人工缺陷制造模块。X~是模拟缺陷的样本，经过EN-DE编码解码器后重建正常样本Y。
测试阶段，X输入EN-DE后得到理想正常样本y，使用LBP对Y和X逐像素特征比较，相差大则有缺陷。
7. MIDL 2018 Unsupervised Detection of Lesions in Brain MRI using constrained adversarial auto-encoders
使用的是AAE来学习建模正常数据分布。有时，对于在正常分布的的两个数据之间的距离，比一个正常和一个异常之间的距离还大，所以提出在隐空间也加一个约束。
暂时先写到这吧。
最后，欢迎关注公众号啦~
最后的最后，再来发一波、到目前为止、部分、用 GAN做异常检测的基本相关（直接用“adversarial anomaly detection”在arxiv上爬下来的，不一定相关！2333）论文供参考！！！
001 (2019-12-10) Event Detection in Micro-PMU Data A Generative Adversarial Network ScoringMethod
arxiv.xilesou.top/pdf/1912.05…
002 (2019-12-10) Outage Detection in Partially Observable DistributionSystems using Smart Meters and Generative Adversarial Networks
arxiv.xilesou.top/pdf/1912.04…
003 (2019-12-9) Oversampling Log Messages Using a Sequence GenerativeAdversarial Network for Anomaly Detection and Classification
arxiv.xilesou.top/pdf/1912.04…
004 (2019-12-2) Anomaly Detection in Particulate Matter Sensor usingHypothesis Pruning Generative Adversarial Network
arxiv.xilesou.top/pdf/1912.00…
005 (2019-11-27) Sparse-GAN Sparsity-constrained Generative Adversarial Network for AnomalyDetection in Retinal OCT Image
arxiv.xilesou.top/pdf/1911.12…
006 (2019-11-21) EvAn NeuromorphicEvent-based Anomaly Detection
arxiv.xilesou.top/pdf/1911.09…
007 (2019-11-19) Attention Guided Anomaly Detection and Localization in Images
arxiv.xilesou.top/pdf/1911.08…
008 (2019-11-17) Deep Verifier Networks Verification of Deep Discriminative Models with Deep Generative Models
arxiv.xilesou.top/pdf/1911.07…
009 (2019-11-16) RSM-GAN A ConvolutionalRecurrent GAN for Anomaly Detection in Contaminated Seasonal Multivariate TimeSeries
arxiv.xilesou.top/pdf/1911.07…
010 (2019-10-30) Robust and Computationally-Efficient Anomaly Detectionusing Powers-of-Two Networks
arxiv.xilesou.top/pdf/1910.14…
011 (2019-10-29) Small-GAN SpeedingUp GAN Training Using Core-sets
arxiv.xilesou.top/pdf/1910.13…
012 (2019-10-23) Photoshopping Colonoscopy Video Frames
arxiv.xilesou.top/pdf/1910.10…
013 (2019-10-21) GraphSAC Detectinganomalies in large-scale graphs
arxiv.xilesou.top/pdf/1910.09…
014 (2019-10-21) Adversarial Anomaly Detection for Marked Spatio-TemporalStreaming Data
arxiv.xilesou.top/pdf/1910.09…
015 (2019-10-10) Misbehaviour Prediction for Autonomous Driving Systems
arxiv.xilesou.top/pdf/1910.04…
016 (2019-10-9) Adversarial Learning of Deepfakes in Accounting
arxiv.xilesou.top/pdf/1910.03…
017 (2019-09-12) Perceptual Image Anomaly Detection
arxiv.xilesou.top/pdf/1909.05…
018 (2019-08-27) Self-Supervised Representation Learning viaNeighborhood-Relational Encoding
arxiv.xilesou.top/pdf/1908.10…
019 (2019-08-10) Transcriptional Response of SK-N-AS Cells to Methamidophos
arxiv.xilesou.top/pdf/1908.03…
020 (2019-09-3) Februus InputPurification Defence Against Trojan Attacks on Deep Neural Network Systems
arxiv.xilesou.top/pdf/1908.03…
021 (2019-08-8) What goes around comes around Cycle-Consistency-based Short-Term MotionPrediction for Anomaly Detection using Generative Adversarial Networks
arxiv.xilesou.top/pdf/1908.03…
022 (2019-08-2) Detection of Accounting Anomalies in the Latent Space usingAdversarial Autoencoder Neural Networks
arxiv.xilesou.top/pdf/1908.00…
023 (2019-09-3) Q-MIND DefeatingStealthy DoS Attacks in SDN with a Machine-learning based Defense Framework
arxiv.xilesou.top/pdf/1907.11…
024 (2019-10-8) Real-time Evasion Attacks with Physical Constraints on DeepLearning-based Anomaly Detectors in Industrial Control Systems
arxiv.xilesou.top/pdf/1907.07…
025 (2019-07-12) AMAD AdversarialMultiscale Anomaly Detection on High-Dimensional and Time-Evolving CategoricalData
arxiv.xilesou.top/pdf/1907.06…
026 (2019-06-27) A Survey on GANs for Anomaly Detection
arxiv.xilesou.top/pdf/1906.11…
027 (2019-06-15) Physical Integrity Attack Detection of Surveillance Camerawith Deep Learning Based Video Frame Interpolation
arxiv.xilesou.top/pdf/1906.06…
028 (2019-07-8) GAN-based Multiple Adjacent Brain MRI Slice Reconstructionfor Unsupervised Alzheimer's Disease Diagnosis
arxiv.xilesou.top/pdf/1906.06…
029 (2019-06-3) Generative Adversarial Networks for Distributed IntrusionDetection in the Internet of Things
arxiv.xilesou.top/pdf/1906.00…
030 (2019-11-20) Unsupervised Learning of Anomaly Detection fromContaminated Image Data using Simultaneous Encoder Training
arxiv.xilesou.top/pdf/1905.11…
031 (2019-10-18) Adversarially-trained autoencoders for robust unsupervisednew physics searches
arxiv.xilesou.top/pdf/1905.10…
032 (2019-05-19) Spatio-Temporal Adversarial Learning for Detecting UnseenFalls
arxiv.xilesou.top/pdf/1905.07…
033 (2019-05-20) Finding Rats in Cats Detecting Stealthy Attacks using Group Anomaly Detection
arxiv.xilesou.top/pdf/1905.07…
034 (2019-04-25) End-to-End Adversarial Learning for Intrusion Detection inComputer Networks
arxiv.xilesou.top/pdf/1904.11…
035 (2019-04-24) GAN Augmented Text Anomaly Detection with Sequences of DeepStatistics
arxiv.xilesou.top/pdf/1904.11…
036 (2019-04-23) A Comparison Study of Credit Card Fraud Detection Supervised versus Unsupervised
arxiv.xilesou.top/pdf/1904.10…
037 (2019-09-24) Trick or Heat Manipulating Critical Temperature-Based Control Systems UsingRectification Attacks
arxiv.xilesou.top/pdf/1904.07…
038 (2019-12-2) Adversarial Learning in Statistical Classification A Comprehensive Review of Defenses AgainstAttacks
arxiv.xilesou.top/pdf/1904.06…
039 (2019-04-11) (Martingale) Optimal Transport And Anomaly Detection WithNeural Networks A Primal-dual Algorithm
arxiv.xilesou.top/pdf/1904.04…
040 (2019-07-24) Efficient GAN-based method for cyber-intrusion detection
arxiv.xilesou.top/pdf/1904.02…
041 (2019-04-2) Fence GAN TowardsBetter Anomaly Detection
arxiv.xilesou.top/pdf/1904.01…
042 (2019-03-27) Fundamental Limits of Covert Packet Insertion
arxiv.xilesou.top/pdf/1903.11…
043 (2019-05-20) Deep Generative Design Integration of Topology Optimization and Generative Models
arxiv.xilesou.top/pdf/1903.01…
044 (2019-11-14) adVAE Aself-adversarial variational autoencoder with Gaussian anomaly prior knowledgefor anomaly detection
arxiv.xilesou.top/pdf/1903.00…
045 (2019-07-14) Secure Distributed Dynamic State Estimation in Wide-AreaSmart Grids
arxiv.xilesou.top/pdf/1902.07…
046 (2019-02-19) Anomaly Detection with Adversarial Dual Autoencoders
arxiv.xilesou.top/pdf/1902.06…
047 (2019-05-9) The Odds are Odd AStatistical Test for Detecting Adversarial Examples
arxiv.xilesou.top/pdf/1902.04…
048 (2019-11-6) BIVA A Very DeepHierarchy of Latent Variables for Generative Modeling
arxiv.xilesou.top/pdf/1902.02…
049 (2019-01-28) Heartbeat Anomaly Detection using Adversarial Oversampling
arxiv.xilesou.top/pdf/1901.09…
050 (2019-01-25) Skip-GANomaly SkipConnected and Adversarially Trained Encoder-Decoder Anomaly Detection
arxiv.xilesou.top/pdf/1901.08…
051 (2019-05-27) Maximum Entropy Generators for Energy-Based Models
arxiv.xilesou.top/pdf/1901.08…
052 (2019-01-10) Adversarial Pseudo Healthy Synthesis Needs PathologyFactorization
arxiv.xilesou.top/pdf/1901.07…
053 (2019-01-18) Robust Anomaly Detection in Images using AdversarialAutoencoders
arxiv.xilesou.top/pdf/1901.06…
054 (2019-01-15) MAD-GAN MultivariateAnomaly Detection for Time Series Data with Generative Adversarial Networks
arxiv.xilesou.top/pdf/1901.04…
055 (2019-12-4) Event Generation and Statistical Sampling for Physics withDeep Generative Models and a Density Information Buffer
arxiv.xilesou.top/pdf/1901.00…
056 (2018-12-11) Anomaly Generation using Generative Adversarial Networks inHost Based Intrusion Detection
arxiv.xilesou.top/pdf/1812.04…
057 (2018-12-11) Anomaly detection with Wasserstein GAN
arxiv.xilesou.top/pdf/1812.02…
058 (2018-12-5) Adversarially Learned Anomaly Detection
arxiv.xilesou.top/pdf/1812.02…
059 (2018-11-11) Adversarial Learning-Based On-Line Anomaly Monitoring forAssured Autonomy
arxiv.xilesou.top/pdf/1811.04…
060 (2018-10-19) Subset Scanning Over Neural Network Activations
arxiv.xilesou.top/pdf/1810.08…
061 (2018-10-11) MDGAN BoostingAnomaly Detection Using \\Multi-Discriminator Generative Adversarial Networks
arxiv.xilesou.top/pdf/1810.05…
062 (2019-04-30) Prospect Theoretic Approach for Data Integrity in IoTNetworks under Manipulation Attacks
arxiv.xilesou.top/pdf/1809.07…
063 (2019-01-15) Anomaly Detection with Generative Adversarial Networks forMultivariate Time Series
arxiv.xilesou.top/pdf/1809.04…
064 (2018-09-28) Layerwise Perturbation-Based Adversarial Training for HardDrive Health Degree Prediction
arxiv.xilesou.top/pdf/1809.04…
065 (2018-09-7) Coupled IGMM-GANs for deep multimodal anomaly detection inhuman mobility data
arxiv.xilesou.top/pdf/1809.02…
066 (2019-08-2) Detection and Mitigation of Attacks on TransportationNetworks as a Multi-Stage Security Game
arxiv.xilesou.top/pdf/1808.08…
067 (2018-08-23) DOPING GenerativeData Augmentation for Unsupervised Anomaly Detection with GAN
arxiv.xilesou.top/pdf/1808.07…
068 (2018-08-1) Anomaly Detection via Minimum Likelihood GenerativeAdversarial Networks
arxiv.xilesou.top/pdf/1808.00…
069 (2018-07-22) SAIFE UnsupervisedWireless Spectrum Anomaly Detection with Interpretable Features
arxiv.xilesou.top/pdf/1807.08…
070 (2018-06-27) Adversarial Distillation of Bayesian Neural NetworkPosteriors
arxiv.xilesou.top/pdf/1806.10…
071 (2019-03-25) Learning Neural Random Fields with Inclusive AuxiliaryGenerators
arxiv.xilesou.top/pdf/1806.00…
072 (2018-07-17) AVID AdversarialVisual Irregularity Detection
arxiv.xilesou.top/pdf/1805.09…
073 (2018-11-13) GANomaly Semi-Supervised Anomaly Detection via Adversarial Training
arxiv.xilesou.top/pdf/1805.06…
074 (2018-05-5) Population Anomaly Detection through Deep Gaussianization
arxiv.xilesou.top/pdf/1805.02…
075 (2018-04-13) Group Anomaly Detection using Deep Generative Models
arxiv.xilesou.top/pdf/1804.04…
076 (2018-04-13) Adversarial Clustering A Grid Based Clustering Algorithm Against Active Adversaries
arxiv.xilesou.top/pdf/1804.04…
077 (2018-04-12) Deep Autoencoding Models for Unsupervised AnomalySegmentation in Brain MR Images
arxiv.xilesou.top/pdf/1804.04…
078 (2018-04-3) Correlated discrete data generation using adversarialtraining
arxiv.xilesou.top/pdf/1804.00…
079 (2018-03-17) A Multi-perspective Approach To Anomaly Detection ForSelf-aware Embodied Agents
arxiv.xilesou.top/pdf/1803.06…
080 (2018-04-9) CIoTA CollaborativeIoT Anomaly Detection via Blockchain
arxiv.xilesou.top/pdf/1803.03…
081 (2018-05-24) Adversarially Learned One-Class Classifier for NoveltyDetection
arxiv.xilesou.top/pdf/1802.09…
082 (2019-05-1) Efficient GAN-Based Anomaly Detection
arxiv.xilesou.top/pdf/1802.06…
083 (2018-02-13) Satellite Image Forgery Detection and Localization UsingGAN and One-Class Classifier
arxiv.xilesou.top/pdf/1802.04…
084 (2018-02-8) Detection of Adversarial Training Examples in PoisoningAttacks through Anomaly Detection
arxiv.xilesou.top/pdf/1802.03…
085 (2018-01-5) Shielding Google's language toxicity model againstadversarial attacks
arxiv.xilesou.top/pdf/1801.01…
086 (2018-06-27) When Not to Classify Anomaly Detection of Attacks (ADA) on DNN Classifiers at Test Time
arxiv.xilesou.top/pdf/1712.06…
087 (2018-04-24) Bayesian Hypernetworks
arxiv.xilesou.top/pdf/1710.04…
088 (2017-09-15) To Go or Not To Go ANear Unsupervised Learning Approach For Robot Navigation
arxiv.xilesou.top/pdf/1709.05…
089 (2017-04-5) Counter-RAPTOR Safeguarding Tor Against Active Routing Attacks
arxiv.xilesou.top/pdf/1704.00…
090 (2017-03-17) Unsupervised Anomaly Detection with Generative AdversarialNetworks to Guide Marker Discovery