Matan Levi

Matan Levi, Aryeh Kontorovich

The existence of adversarial examples points to a basic weakness of deep neural networks. One of the most effective defenses against such examples, adversarial training, entails training models with some degree of robustness, usually at the expense of a degraded natural accuracy. Most adversarial training methods aim to learn a model that finds, for each class, a common decision boundary encompassing both the clean and perturbed examples. In this work, we take a fundamentally different approach by treating the perturbed examples of each class as a separate class to be learned, effectively splitting each class into two classes: "clean" and "adversarial." This split doubles the number of classes to be learned, but at the same time considerably simplifies the decision boundaries. We provide a theoretical plausibility argument that sheds some light on the conditions under which our approach can be expected to be beneficial. Likewise, we empirically demonstrate that our method learns robust models while attaining optimal or near-optimal natural accuracy, e.g., on CIFAR-10 we obtain near-optimal natural accuracy of $95.01\%$ alongside significant robustness across multiple tasks. The ability to achieve such near-optimal natural accuracy, while maintaining a significant level of robustness, makes our method applicable to real-world applications where natural accuracy is at a premium. As a whole, our main contribution is a general method that confers a significant level of robustness upon classifiers with only minor or negligible degradation of their natural accuracy.

Matan Levi, Idan Attias, Aryeh Kontorovich

The phenomenon of adversarial examples illustrates one of the most basic vulnerabilities of deep neural networks. Among the variety of techniques introduced to surmount this inherent weakness, adversarial training has emerged as the most effective strategy for learning robust models. Typically, this is achieved by balancing robust and natural objectives. In this work, we aim to further optimize the trade-off between robust and standard accuracy by enforcing a domain-invariant feature representation. We present a new adversarial training method, Domain Invariant Adversarial Learning (DIAL), which learns a feature representation that is both robust and domain invariant. DIAL uses a variant of Domain Adversarial Neural Network (DANN) on the natural domain and its corresponding adversarial domain. In the case where the source domain consists of natural examples and the target domain is the adversarially perturbed examples, our method learns a feature representation constrained not to discriminate between the natural and adversarial examples, and can therefore achieve a more robust representation. DIAL is a generic and modular technique that can be easily incorporated into any adversarial training method. Our experiments indicate that incorporating DIAL in the adversarial training process improves both robustness and standard accuracy.

Matan Levi, Yair Allouche, Aryeh Kontorovich

The vehicular connectivity revolution is fueling the automotive industry's most significant transformation seen in decades. However, as modern vehicles become more connected, they also become much more vulnerable to cyber-attacks. In this paper, a fully working machine learning approach is proposed to protect connected vehicles (fleets and individuals) against such attacks. We present a system that monitors different vehicle's interfaces (Network, CAN and OS), extracts relevant information based on configurable rules and sends it to a trained generative model to detect deviations from normal behavior. Using configurable data collector, we provide a higher level of data abstraction as the model is trained based on events instead of raw data, which has a noise-filtering effect and eliminates the need to retrain the model whenever a protocol changes. We present a new approach for detecting anomalies, tailored to the temporal nature of our domain. Adapting the hybrid approach of Gutflaish et al. (2017) to the fully temporal setting, we first train a Hidden Markov Model to learn normal vehicle behavior, and then a regression model to calibrate the likelihood threshold for anomaly. Using this architecture, our method detects sophisticated and realistic anomalies, which are missed by other existing methods monitoring the CAN bus only. We also demonstrate the superiority of adaptive thresholds over static ones. Furthermore, our approach scales efficiently from monitoring individual cars to serving large fleets. We demonstrate the competitive advantage of our model via encouraging empirical results.