PAD: Towards Principled Adversarial Malware Detection Against Evasion Attacks

Machine Learning (ML) techniques facilitate automating malicious software (malware for short) detection, but suffer from evasion attacks. Many researchers counter such attacks in heuristic manners short of both theoretical guarantees and defense effectiveness. We hence propose a new adversarial training framework, termed Principled Adversarial Malware Detection (PAD), which encourages convergence guarantees for robust optimization methods. PAD lays on a learnable convex measurement that quantifies distribution-wise discrete perturbations and protects the malware detector from adversaries, by which for smooth detectors, adversarial training can be performed heuristically with theoretical treatments. To promote defense effectiveness, we propose a new mixture of attacks to instantiate PAD for enhancing the deep neural network-based measurement and malware detector. Experimental results on two Android malware datasets demonstrate: (i) the proposed method significantly outperforms the state-of-the-art defenses; (ii) it can harden the ML-based malware detection against 27 evasion attacks with detection accuracies greater than 83.45%, while suffering an accuracy decrease smaller than 2.16% in the absence of attacks; (iii) it matches or outperforms many anti-malware scanners in VirusTotal service against realistic adversarial malware.

Can We Leverage Predictive Uncertainty to Detect Dataset Shift and Adversarial Examples in Android Malware Detection?

The deep learning approach to detecting malicious software (malware) is promising but has yet to tackle the problem of dataset shift, namely that the joint distribution of examples and their labels associated with the test set is different from that of the training set. This problem causes the degradation of deep learning models without users' notice. In order to alleviate the problem, one approach is to let a classifier not only predict the label on a given example but also present its uncertainty (or confidence) on the predicted label, whereby a defender can decide whether to use the predicted label or not. While intuitive and clearly important, the capabilities and limitations of this approach have not been well understood. In this paper, we conduct an empirical study to evaluate the quality of predictive uncertainties of malware detectors. Specifically, we re-design and build 24 Android malware detectors (by transforming four off-the-shelf detectors with six calibration methods) and quantify their uncertainties with nine metrics, including three metrics dealing with data imbalance. Our main findings are: (i) predictive uncertainty indeed helps achieve reliable malware detection in the presence of dataset shift, but cannot cope with adversarial evasion attacks; (ii) approximate Bayesian methods are promising to calibrate and generalize malware detectors to deal with dataset shift, but cannot cope with adversarial evasion attacks; (iii) adversarial evasion attacks can render calibration methods useless, and it is an open problem to quantify the uncertainty associated with the predicted labels of adversarial examples (i.e., it is not effective to use predictive uncertainty to detect adversarial examples).

Adversarial Deep Ensemble: Evasion Attacks and Defenses for Malware Detection

Malware remains a big threat to cyber security, calling for machine learning based malware detection. While promising, such detectors are known to be vulnerable to evasion attacks. Ensemble learning typically facilitates countermeasures, while attackers can leverage this technique to improve attack effectiveness as well. This motivates us to investigate which kind of robustness the ensemble defense or effectiveness the ensemble attack can achieve, particularly when they combat with each other. We thus propose a new attack approach, named mixture of attacks, by rendering attackers capable of multiple generative methods and multiple manipulation sets, to perturb a malware example without ruining its malicious functionality. This naturally leads to a new instantiation of adversarial training, which is further geared to enhancing the ensemble of deep neural networks. We evaluate defenses using Android malware detectors against 26 different attacks upon two practical datasets. Experimental results show that the new adversarial training significantly enhances the robustness of deep neural networks against a wide range of attacks, ensemble methods promote the robustness when base classifiers are robust enough, and yet ensemble attacks can evade the enhanced malware detectors effectively, even notably downgrading the VirusTotal service.

SoK: Arms Race in Adversarial Malware Detection

Malicious software (malware) is a major cyber threat that shall be tackled with Machine Learning (ML) techniques because millions of new malware examples are injected into cyberspace on a daily basis. However, ML is known to be vulnerable to attacks known as adversarial examples. In this SoK paper, we systematize the field of Adversarial Malware Detection (AMD) through the lens of a unified framework of assumptions, attacks, defenses and security properties. This not only guides us to map attacks and defenses into some partial order structures, but also allows us to clearly describe the attack-defense arms race in the AMD context. In addition to manually drawing insights, we also propose using ML to draw insights from the systematized representation of the literature. Examples of the insights are: knowing the defender's feature set is critical to the attacker's success; attack tactic (as a core part of the threat model) largely determines what security property of a malware detector can be broke; there is currently no silver bullet defense against evasion attacks or poisoning attacks; defense tactic largely determines what security properties can be achieved by a malware detector; knowing attacker's manipulation set is critical to defender's success; ML is an effective method for insights learning in SoK studies. These insights shed light on future research directions.

Enhancing Deep Neural Networks Against Adversarial Malware Examples

Machine learning based malware detection is known to be vulnerable to adversarial evasion attacks. The state-of-the-art is that there are no effective countermeasures against these attacks. Inspired by the AICS'2019 Challenge organized by the MIT Lincoln Lab, we systematize a number of principles for enhancing the robustness of neural networks against adversarial malware evasion attacks. Some of these principles have been scattered in the literature, but others are proposed in this paper for the first time. Under the guidance of these principles, we propose a framework for defending against adversarial malware evasion attacks. We validated the framework using the Drebin dataset of Android malware. We applied the defense framework to the AICS'2019 Challenge and won, without knowing how the organizers generated the adversarial examples. However, we see a ~22\% difference between the accuracy in the experiment with the Drebin dataset (for binary classification) and the accuracy in the experiment with respect to the AICS'2019 Challenge (for multiclass classification). We attribute this gap to a fundamental barrier that without knowing the attacker's manipulation set, the defender cannot do effective Adversarial Training.

HashTran-DNN: A Framework for Enhancing Robustness of Deep Neural Networks against Adversarial Malware Samples

Adversarial machine learning in the context of image processing and related applications has received a large amount of attention. However, adversarial machine learning, especially adversarial deep learning, in the context of malware detection has received much less attention despite its apparent importance. In this paper, we present a framework for enhancing the robustness of Deep Neural Networks (DNNs) against adversarial malware samples, dubbed Hashing Transformation Deep Neural Networks} (HashTran-DNN). The core idea is to use hash functions with a certain locality-preserving property to transform samples to enhance the robustness of DNNs in malware classification. The framework further uses a Denoising Auto-Encoder (DAE) regularizer to reconstruct the hash representations of samples, making the resulting DNN classifiers capable of attaining the locality information in the latent space. We experiment with two concrete instantiations of the HashTran-DNN framework to classify Android malware. Experimental results show that four known attacks can render standard DNNs useless in classifying Android malware, that known defenses can at most defend three of the four attacks, and that HashTran-DNN can effectively defend against all of the four attacks.