[論文レビュー] Efficient GAN-Based Anomaly Detection

Houssam Zenati, Chuan-Sheng Foo+3 more|arXiv (Cornell University)|Feb 17, 2018

Anomaly Detection Techniques and Applications参考文献 16被引用数 483

ひとこと要約

この論文は BiGAN に似たフレームワークを用い、エンコーダを組み込んだ効率的な異常検知を実現し、MNISTとKDD99で最先端の結果を、従来のGANベース手法よりもはるかに高速なテスト時推論とともに達成している。

ABSTRACT

Generative adversarial networks (GANs) are able to model the complex highdimensional distributions of real-world data, which suggests they could be effective for anomaly detection. However, few works have explored the use of GANs for the anomaly detection task. We leverage recently developed GAN models for anomaly detection, and achieve state-of-the-art performance on image and network intrusion datasets, while being several hundred-fold faster at test time than the only published GAN-based method.

研究の動機と目的

• モデル normal data distribution with a GAN that jointly learns an encoder and a generator.
• Avoid costly latent-embedding recovery at test time by integrating encoding into training.
• Define an anomaly score combining reconstruction loss and discriminator-based loss.
• Demonstrate state-of-the-art performance on image (MNIST) and network intrusion (KDD99) datasets.

提案手法

• BiGAN-style training を採用して正規データの学習のために G, E, D を同時に学習する。
• min_G,E max_D の損失 V(D,E,G) を最適化して x を z に合わせる。
• 異常スコア A(x) = α L_G(x) + (1−α) L_D(x) を定義し、L_G(x) = ||x − G(E(x))||_1。
• L_D の2つのバリアントを評価する：クロスエントロピー σ(D(x,E(x)),1) と特徴量一致 ||f_D(x,E(x)) − f_D(G(E(x)),E(x))||_1。
• 特徴量マッチング L_D (FM) が異常スコアリングでクロスエントロピー σ より性能を向上させることを示す。
• AnoGAN および VAE と比較し、MNIST と KDD99 の結果と推論時の速度向上を報告する。

実験結果

リサーチクエスチョン

• RQ1 Can a GAN with an encoder trained jointly with the generator perform efficient anomaly detection without test-time latent recovery?
• RQ2 Does incorporating an encoder into GAN training improve anomaly detection on high-dimensional data such as images and network traffic?
• RQ3 How does the BiGAN-based anomaly detector compare to existing GAN-based and non-GAN methods on MNIST and KDD99 in terms of accuracy and speed?
• RQ4 What is the impact of using a feature-matching discriminator loss versus a cross-entropy discriminator loss in anomaly scoring?

主な発見

Model	Precision	Recall	F1
OC-SVM	0.7457	0.8523	0.7954
DSEBM-r	0.8521	0.6472	0.7328
DSEBM-e	0.8619	0.6446	0.7399
DAGMM-NVI	0.9290	0.9447	0.9368
DAGMM	0.9297	0.9442	0.9369
AnoGAN FM	0.8698	0.9523	0.9058
AnoGAN σ	0.7790	0.7914	0.7852
Our Model FM	0.8698	0.9523	0.9058
Our Model σ	0.9200	0.9582	0.9372

• On MNIST, the BiGAN-based method outperforms AnoGAN and VAE in AUPRC across 10 MNIST-derived anomaly tasks.
• The FM variant of L_D yields better anomaly scores than the σ variant, consistent with discriminator features being informative for detection.
• Inference-time speed is dramatically faster than AnoGAN (approx. 800x faster on MNIST).
• On KDD99, the method remains competitive with state-of-the-art results and achieves higher recall, while also delivering substantial speedups (700x to 900x faster).
• AnoGAN-FM and the BiGAN-based method show strong cross-dataset generalization for high-dimensional data.