Weakly-supervised Anomaly Detection: A Survey

This repo is constructed for collecting and categorizing papers about weakly supervised anomaly detection models according to our survey paper——Weakly Supervised Anomaly Detection: A Survey

Summary and categorization of weakly supervised anomaly detection (WSAD) algorithms

We first summarize and further categorize existing WSAD algorithms into three categories, including: (i) incomplete supervision; (ii) inexact supervision; (iii) inaccurate supervision

1.Summary of WSAD Algorithms

Method	Reference	Venue	Backbone	Modalities	Key Idea	Official Code
			Incomplete Supervision
OE	ref	KDD'14	-	Tabular	Anomaly feature representation learning	×
XGBOD	ref	IJCNN'18	-	Tabular	Anomaly feature representation learning	√
DeepSAD	ref	ICLR'20	MLP	Tabular	Anomaly feature representation learning	√
ESAD	ref	Preprint	MLP	Tabular	Anomaly feature representation learning	×
DSSAD	ref	ICASSP'21	CNN	Image/Video	Anomaly feature representation learning	×
REPEN	ref	KDD'18	MLP	Tabular	Anomaly feature representation learning	×
AA-BiGAN	ref	IJCAI'22	GAN	Tabular	Anomaly feature representation learning	√
Dual-MGAN	ref	TKDD'22	GAN	Tabular	Anomaly feature representation learning	√
DevNet	ref	KDD'19	MLP	Tabular	Anomaly score learning	√
PReNet	ref	Preprint	MLP	Tabular	Anomaly score learning	×
FEAWAD	ref	TNNLS'21	AE	Tabular	Anomaly score learning	√
SNARE	ref	KDD'09	-	Graph	Graph learning and label propagation	×
AESOP	ref	KDD'14	-	Graph	Graph learning and label propagation	×
SemiGNN	ref	ICDM'19	MLP+Attention	Graph	Graph learning and label propagation	×
SemiGAD	ref	IJCNN'21	GNN	Graph	Graph learning and label propagation	×
Meta-GDN	ref	WWW'21	GNN	Graph	Graph learning and label propagation	√
SemiADC	ref	IS Journal'21	GAN	Graph	Graph learning and label propagation	×
SSAD	ref	JAIR'13	-	Tabular	Active learning	×
AAD	ref	ICDM'16	-	Tabular	Active learning	√
SLA-VAE	ref	WWW'22	VAE	Time series	Active learning	×
Meta-AAD	ref	ICDM'20	MLP	Tabular	Reinforcement learning	√
DPLAN	ref	KDD'21	MLP	Tabular	Reinforcement learning	×
GraphUCB	ref	WSDM'19	-	Graph	Reinforcement learning	√
			Inexact Supervision
MIL	ref	CVPR'18	MLP	Video	Multiple Instance Learning	√
TCN-IBL	ref	ICIP'19	CNN	Video	Multiple Instance Learning	×
AR-Net	ref	ICME'20	MLP	Video	Multiple Instance Learning	√
RTFM	ref	ICCV'21	CNN+Attention	Video	Multiple Instance Learning	√
Motion-Aware	ref	BMVC'19	AE+Attention	Video	Multiple Instance Learning	×
CRF-Attention	ref	ICCV'21	TRN+Attention	Video	Multiple Instance Learning	×
MPRF	ref	IJCAI'21	MLP+Attention	Video	Multiple Instance Learning	×
MCR	ref	ICME'22	MLP+Attention	Video	Multiple Instance Learning	×
XEL	ref	SPL'21	MLP	Video	Cross-epoch Learning	√
MIST	ref	CVPR'21	MLP+Attention	Video	Multiple Instance Learning	√
MSLNet	ref	AAAI'22	Transformer	Video	Multiple Instance Learning	√
SRF	ref	SPL'20	MLP	Video	Self Reasoning	×
WETAS	ref	ICCV'21	MLP	Time-series/Video	Dynamic Time Warping	×
Inexact AUC	ref	ML Journal'20	AE	Tabular	AUC maximization	×
Isudra	ref	TIST'21	-	Time-series	Bayesian optimization	√
			Inaccurate Supervision
LAC	ref	CIKM'21	MLP/GBDT	Tabular	Ensemble learning	×
ADMoE	ref	AAAI'23	Agnostic	Tabular	Ensemble learning	√
BGPAD	ref	ICNP'21	LSTM+Attention	Time series	Denoising network	√
SemiADC	ref	IS Journal'21	GAN	Graph	Denoising network	×
TSN	ref	CVPR'19	GCN	Video	GCN	√

2.Categorization of WSAD algorithms

2.1 AD with Incomplete Supervision

• Anomaly Feature Representation Learning

  • OE
    📄Learning outlier ensembles:The best of both worlds–supervised and unsupervised\
  • XGBOD
    📄Xgbod: improving supervised outlier detection with unsupervised representation learning
    👉Code Link
  • DeepSAD
    📄Deep semi-supervised anomaly detection
    👉Code Link
  • ESAD
    📄Esad: End-to-end deep semi-supervised anomaly detection
  • REPEN
    📄Learning representations of ultrahigh-dimensional data for random distance-based outlier detection
  • DSSAD
    📄Learning discriminative features for semi-supervised anomaly detection
  • AA-BiGAN
    📄Anomaly detection by leveraging incomplete anomalous knowledge with anomaly-aware bidirectional gans
    👉Code Link
  • Dual-MGAN
    📄Dual-mgan: An efficient approach for semi-supervised outlier detection with few identified anomalies
    👉Code Link

• Anomaly Score Learning

  • DevNet
    📄Deep anomaly detection with deviation networks
    👉Code Link
  • PReNet
    📄Deep weakly-supervised anomaly detection
  • FEAWAD
    📄Feature encoding with autoencoders for weakly supervised anomaly detection
    👉Code Link

• Graph Learning

  • SNARE
    📄Snare: a link analytic system for graph labeling and risk detection
  • AESOP
    📄Guilt by association: large scale malware detection by mining file-relation graphs
  • SemiGNN
    📄A semi-supervised graph attentive network for financial fraud detection
  • SemiGAD
    📄Semi-supervised anomaly detection on attributed graphs
  • Meta-GDN
    📄Few-shot network anomaly detection via cross-network meta-learning
    👉Code Link
  • SemiADC
    📄Semi-supervised anomaly detection in dynamic communication networks
  • SSAD
    📄Toward supervised anomaly detection
  • AAD
    📄Incorporating expert feedback into active anomaly discover
    👉Code Link
  • GraphUCB
    📄Interactive anomaly detection on attributed networks
    👉Code Link

• Active learning and reinforcement learning

  • Meta-AAD
    📄Meta-aad: Active anomaly detection with deep reinforcement learning
    👉Code Link
  • DPLAN
    📄Toward deep supervised anomaly detection: Reinforcement learning from partially labeled anomaly data
  • SLA-VAE
    📄A semi-supervised vae based active anomaly detection framework in multivariate time series for online systems

2.2 AD with Inexact Supervision

• MIL-based
  • MIL
    📄Real-world anomaly detection in surveillance videos
    👉Code Link
  • AR-Net
    📄Weakly supervised video anomaly detection via center-guided discriminative learning
    👉Code Link
  • TCN-IBL
    📄Temporal convolutional network with complementary inner bag loss for weakly supervised anomaly detection
  • RTFM
    📄Weakly-supervised video anomaly detection with robust temporal feature magnitude learning
    👉Code Link
  • Motion-Aware
    📄Motion-aware feature for improved video anomaly detection
  • CRF-Attention
    📄Dance with self-attention: A new look of conditional random fields on anomaly detection in videos
  • MPRF
    📄Weakly-supervised spatio-temporal anomaly detection in surveillance video
  • MCR
    📄Multi-scale continuity-aware refinement network for weakly supervised video anomaly detection
  • XEL
    📄Cross-epoch learning for weakly supervised anomaly detection in surveillance videos
    👉Code Link
  • MIST
    📄MIST: Multiple instance self-training framework for video anomaly detection
    👉Code Link
  • MSLNet
    📄Self-training multi-sequence learning with transformer for weakly supervised video anomaly detection
    👉Code Link
• Non MIL-based
  • SRF
    📄A self-reasoning framework for anomaly detection using video-level labels
  • WETAS
    📄Weakly supervised temporal anomaly seg- mentation with dynamic time warping
• Evaluating and Selecting Unsupervised methods
  • Inexact AUC
    📄Anomaly detection with inexact labels
  • Isudra
    📄Indirectly supervised anomaly detection of clinically meaningful health events from smart home data
    👉Code Link

2.3 AD with Inaccurate Supervision

• Ensemble Learning

  • LAC
    📄Fraud Detection under Multi-Sourced Extremely Noisy Annotations
  • ADMoE
    📄ADMoE: Anomaly detection with mixture-of-experts from noisy labels
    👉Code Link

• Denosing Network

  • BGPAD
    📄Isp self-operated bgp anomaly detection based on weakly supervised learning
    👉Code Link
  • SemiADC
    📄Semi-supervised anomaly detection in dynamic communication networks

• Graph Learning

  • TSN
    📄Graph Convolutional Label Noise Cleaner: Train a Plug-And-Play Action Classifier for Anomaly Detection
    👉Code Link

Experiment

One can easily reproduce the experimental results in our paper by running the run.py python file in the experiments folder.

Method	$\gamma_{l}=1$%	$\gamma_{l}=5$%	$\gamma_{l}=25$%	$\gamma_{l}=50$%
		AUC-ROC
XGBOD	80.03	86.68	93.20	95.28
DeepSAD	75.25	81.74	89.64	92.72
REPEN	77.20	82.23	86.26	87.45
DevNet	79.05	85.94	89.76	90.97
PReNet	79.04	85.66	89.88	91.11
FEAWAD	73.93	82.44	89.20	91.55
		AUC-PR
XGBOD	46.23	61.58	75.89	80.57
DeepSAD	38.06	49.65	67.04	74.47
REPEN	46.57	56.38	63.39	65.73
DevNet	53.61	64.01	69.52	71.13
PReNet	54.52	64.19	70.46	71.62
FEAWAD	51.19	62.30	69.65	72.34