EVAGAN: Evasion Generative Adversarial Network for Low Data Regimes

Abstract

Many recent literary works have leveraged generative adversarial networks (GANs) to spawn unseen evasion samples. The purpose is to annex the generated data with the original train set for adversarial training to improve the detection performance of machine learning (ML) classifiers. The quality of generating adversarial samples relies on the adequacy of training data samples. However, in low data regimes like medical anomaly detection, drug discovery and cybersecurity, the attack samples are scarce in number. This paper proposes a novel GAN design called Evasion Generative Adversarial Network (EVAGAN) that is more suitable for low data regime problems that use oversampling for detection improvement of ML classifiers. EVAGAN not only can generate evasion samples, but its discriminator can act as an evasion aware classifier. We have considered Auxiliary Classifier GAN (ACGAN) as a benchmark to evaluate the performance of EVAGAN on cybersecurity (ISCX-2014, CIC-2017 and CIC2018) botnet and CV (MNIST) datasets. We demonstrate that EVAGAN outperforms ACGAN for unbalanced datasets with respect to detection performance, training stability, time complexity. EVAGAN's generator quickly learns to generate the low sample class and hardens its discriminator simultaneously. In contrast to ML classifiers that require security hardening after being adversarially trained by GAN generated data, EVAGAN renders it needless. The experimental analysis proves EVAGAN to be an efficient evasion hardened model for low data regimes in cybersecurity and CV. Code will be available at https://github.com/rhr407/EVAGAN.Impact Statement-The applications of Artificial Intelligence (AI) can help improve the quality of human life. The use of AI is not only limited to medical anomaly detection and drug discovery but can be leveraged in computer networks to keep people safe from malicious activities on the Internet. However, the AI-based models can be biased towards the majority class of data on which they are trained due to data imbalance. Anomaly data samples are always scarce as compared to the normal data samples. So this is an open research problem to solve. Our work is an effort to improve the AI-based methods in detection performance, time complexity and stability. Using the proposed technique, we can train our AI model using fewer anomaly samples efficiently and reduce the time complexity compared to the state-of-the-art in anomaly detection.

Deep Reinforcement Learning based Evasion Generative Adversarial Network for Botnet Detection

“…These estimations were computed using the Keras model.predict function. The details of these metrics can be found in EVAGAN paper [28]. Figure 5 illustrates the losses of D for real and fake minority classes and majority/normal classes and of G for ACGAN, EVAGAN and RELEVAGAN.…”

“…One is EVAGAN, summarised in Section II-E, and the other part is a DRL-based model. For more details on EVAGAN, readers are highly encouraged to refer to the paper [28]. A typical DRL model consisting of an agent and an environment has been coupled with EVAGAN architecture in RELEVAGAN design.…”

“…Table II shows the distribution of benign and botnet samples in all three datasets. The details of a particular botnet selection are mentioned in EVAGAN paper [28].…”

Deep Reinforcement Learning Based Evasion Generative Adversarial Network for Botnet Detection

“…However, extending the capability of D of a GAN from discriminating between real and fake samples to differentiating between normal and anomaly data renders the adversarial training needless because D can act as an evasion-aware classifier [17,18]. In [19], authors propose EVAGAN that provides such type of D and compare its performance with the D of ACGAN and other ML classifiers like xgboost (XGB), naive bayes (NB), decision tree (DT), random forests (RF), knearest neighbours (KNN) and logistic regression (LR). EVAGAN's D outperforms the ML classifiers in black box testing and gives 100% accuracy in normal and evasion sample estimation.…”

“…One is EVAGAN, summarised in Section 2.5, and the other is a DRL-based model. For more details on EVAGAN, readers are encouraged to refer to the paper [19]. A typical DRL model consisting of an agent and an environment has been coupled with EVAGAN architecture in RELEVAGAN design.…”