No abstract

Deep learning (DL) systems are increasingly deployed in safety-and security-critical domains including self-driving cars and malware detection, where the correctness and predictability of a system's behavior for corner case inputs are of great importance. Existing DL testing depends heavily on manually labeled data and therefore often fails to expose erroneous behaviors for rare inputs.We design, implement, and evaluate DeepXplore, the first whitebox framework for systematically testing real-world DL systems. First, we introduce neuron coverage for systematically measuring the parts of a DL system exercised by test inputs. Next, we leverage multiple DL systems with similar functionality as cross-referencing oracles to avoid manual checking. Finally, we demonstrate how finding inputs for DL systems that both trigger many differential behaviors and achieve high neuron coverage can be represented as a joint optimization problem and solved efficiently using gradientbased search techniques.DeepXplore efficiently finds thousands of incorrect corner case behaviors (e.g., self-driving cars crashing into guard rails and malware masquerading as benign software) in stateof-the-art DL models with thousands of neurons trained on five popular datasets including ImageNet and Udacity selfdriving challenge data. For all tested DL models, on average, DeepXplore generated one test input demonstrating incorrect behavior within one second while running only on a commodity laptop. We further show that the test inputs generated by DeepXplore can also be used to retrain the corresponding DL model to improve the model's accuracy by up to 3%.

In this paper, we propose a browser fingerprinting technique that can track users not only within a single browser but also across different browsers on the same machine. Specifically, our approach utilizes many novel OS and hardware level features, such as those from graphics cards, CPU, and installed writing scripts. We extract these features by asking browsers to perform tasks that rely on corresponding OS and hardware functionalities.Our evaluation shows that our approach can successfully identify 99.24% of users as opposed to 90.84% for state of the art on single-browser fingerprinting against the same dataset. Further, our approach can achieve higher uniqueness rate than the only cross-browser approach in the literature with similar stability.In the paper, we propose a (cross-)browser fingerprinting based on many novel OS and hardware level features, e.g., these from graphics card, CPU, audio stack, and installed writing scripts. Specifically, because many of such OS and hardware level functions are exposed to JavaScript via browser APIs, we can extract features when asking the browser to perform certain tasks through these APIs. The extracted features can be used for both single-and cross-browser fingerprinting.

A wealth of recent research proposes static data flow analysis for the security analysis of Android applications. One of the building blocks that these analysis systems rely upon is the computation of a precise control flow graph. The callback mechanism provided and orchestrated by the Android framework makes the correct generation of the control flow graph a challenging endeavor. From the analysis' point of view, the invocation of a callback is an implicit control flow transition facilitated by the framework. Existing static analysis tools model callbacks either through manually-curated lists or ad-hoc heuristics. This work demonstrates that both approaches are insufficient, and allow malicious applications to evade detection by state-of-theart analysis systems. To address the challenge of implicit control flow transitions (i.e., callbacks) through the Android framework, we are the first to propose, implement, and evaluate a systematic treatment of this aspect. Our implementation, called EDGEMINER, statically analyzes the entire Android framework to automatically generate API summaries that describe implicit control flow transitions through the Android framework. We use EDGEMINER to analyze three major versions of the Android framework. EDGEMINER identified 19,647 callbacks in Android 4.2, suggesting that a manual treatment of this challenge is likely infeasible. Our evaluation demonstrates that the current insufficient treatment of callbacks in state-of-the-art analysis tools results in unnecessary imprecision. For example, FlowDroid misses a variety of leaks of privacy sensitive data from benign off-the-shelf Android applications because of its inaccurate handling of callbacks. Of course, malicious applications can also leverage this blind spot in current analysis systems to evade detection at will. The results of our work allow existing tools to comprehensively address the challenge of callbacks and identify previously undetected leakage of privacy sensitive data.

No abstract

No abstract

Due to the increasing usage of machine learning (ML) techniques in security-and safety-critical domains, such as autonomous systems and medical diagnosis, ensuring correct behavior of ML systems, especially for different corner cases, is of growing importance. In this paper, we propose a generic framework for evaluating security and robustness of ML systems using different real-world safety properties. We further design, implement and evaluate VERIVIS, a scalable methodology that can verify a diverse set of safety properties for state-of-the-art computer vision systems with only blackbox access. VERIVIS leverage different input space reduction techniques for efficient verification of different safety properties. VERIVIS is able to find thousands of safety violations in fifteen state-of-the-art computer vision systems including ten Deep Neural Networks (DNNs) such as Inception-v3 and Nvidia's Dave self-driving system with thousands of neurons as well as five commercial third-party vision APIs including Google vision and Clarifai for twelve different safety properties. Furthermore, VERIVIS can successfully verify local safety properties, on average, for around 31.7% of the test images. VERIVIS finds up to 64.8× more violations than existing gradient-based methods that, unlike VERIVIS, cannot ensure non-existence of any violations. Finally, we show that retraining using the safety violations detected by VERIVIS can reduce the average number of violations up to 60.2%.

Membership inference (MI) attacks affect user privacy by inferring whether given data samples have been used to train a target learning model, e.g., a deep neural network. There are two types of MI attacks in the literature, i.e., these with and without shadow models. The success of the former heavily depends on the quality of the shadow model, i.e., the transferability between the shadow and the target; the latter, given only blackbox probing access to the target model, cannot make an effective inference of unknowns, compared with MI attacks using shadow models, due to the insufficient number of qualified samples labeled with ground truth membership information. In this paper, we propose an MI attack, called BLINDMI, which probes the target model and extracts membership semantics via a novel approach, called differential comparison. The high-level idea is that BLINDMI first generates a dataset with nonmembers via transforming existing samples into new samples, and then differentially moves samples from a target dataset to the generated, non-member set in an iterative manner. If the differential move of a sample increases the set distance, BLINDMI considers the sample as non-member and vice versa. BLINDMI was evaluated by comparing it with state-of-theart MI attack algorithms. Our evaluation shows that BLINDMI improves F1-score by nearly 20% when compared to state-ofthe-art on some datasets, such as Purchase-50 and Birds-200, in the blind setting where the adversary does not know the target model's architecture and the target dataset's ground truth labels. We also show that BLINDMI can defeat state-of-the-art defenses.