Z. Berkay Celik336 Westgate Building
University Park, PA 16802, USA
Github Linkedin Twitter
I am a research assistant in Department of Electrical Engineering and Computer Science at the Pennsylvania State University working with Prof. Patrick McDaniel and a member of the Systems and Internet Infrastructure Security Laboratory (SIIS). During my Msc. studies, I worked with Prof. George Kesidis and Prof. David J. Miller. My research was on machine learning systems and network security.
Generally, I am interested in developing secure systems through program analysis and machine learning. My recent work seeks to improve security, safety and privacy guarantees in commodity IoT.
I’ve had the opportunity to work on a number of interesting research projects during my Msc. and PhD. studies. Here is a summary of some of my efforts.
Automated IoT Security, Safety and Privacy Analysis
First, we implemented Soteria, a static analysis system for validating whether an IoT app or IoT environment (collection of apps working in concert) adheres to identified safety, security, and functional properties. We demonstrated that many apps violate properties when used in isolation and when used together in multi-app environments. Read more about Soteria. Second, we present Saint, a static taint analysis tool for IoT applications. Through this effort, we introduced a rigorously grounded framework for evaluating the use of sensitive information in IoT apps–and therein provide developers, markets, and consumers a means of identifying potential threats to security and privacy. Read more about Saint.
IoTBench is an IoT-specific test corpus, an open repository for evaluating systems designed for IoT app analyses. IoTBench is under continual development, always welcoming contributions of the new IoT apps.
Detection under Privileged Information
The training of the models has been historically limited to only those features available at runtime. In this study, we ask “how can detection systems integrate intelligence relevant to an attack that is available at training time, yet not available at runtime?”. Consider a rootkit detection system for mobile phones. Obtaining a set of features may drain the battery even though they contribute to the detection of a rootkit. Therefore, the users might disable the system because the battery dies very quickly. However, obtaining these features at training time is feasible as we have unlimited resources in a laboratory-based environment. In this study, we cover an alternate model construction approach that trains models using “privileged” information–features available at training time but not at runtime, yet classifies new samples without the need of privileged features at runtime. Turning to rootkit example, the features draining the battery quickly are defined privileged features.
Our paper is accepted to AsiaCCS’18: Detection under Privileged Information. Read more about formulation and implementation in our technical report, and Feature Cultivation in Privileged Information-augmented Detection.
Machine Learning in Adversarial Settings
One of the limitations of machine learning in practice is that they are subject to adversarial samples. Adversarial samples are carefully modified inputs crafted to dictate a selected output. In the context of classification, adversarial samples are crafted so as to force a target model to classify them in a class different from their legitimate class. In this work, we focus on Deep Neural Networks (DNNs) for adversarial sample generation and attacker’s capabilities to evade systems built on DNNs. Check out our publication on adversarial machine learning: The Limitations of Deep Learning in Adversarial Settings and Practical Black-Box Attacks against Machine Learning.
Patient-Driven Privacy Control
Patients are asked to disclose personal information such as genetic markers, lifestyle habits, and clinical history. This data is then used by statistical models to predict personalized treatments. However, due to privacy concerns, patients often desire to withhold sensitive information. This self-censorship can impede proper diagnosis and treatment, which may lead to serious health complications. We implemented privacy distillation, a mechanism which allows patients to control the type and amount of information they wish to disclose to the healthcare providers for use in statistical models.
This paper is accepted to IEEE Privacy-aware computing (PAC) conference Patient-Driven Privacy Control through Generalized Distillation, 2016. The use of patient data made us develop new algorithms for Achieving Secure and Differentially Private Computations in Multiparty Settings. This paper is also accepted to PAC conference.
Malware Traffic Detection and Malware Experimentation Design
We present a framework for evaluating the transport layer feature space of malware heartbeat traffic. We use these features in a prototype detection system to distinguish malware traffic from traffic generated by legitimate applications. Further, we characterize the evolution of malware evasion techniques over time by examining the behavior of 16 malware families. In particular, we highlight the difficulty of detecting malware that use traffic-shaping techniques to mimic legitimate traffic. Read more about the study here.
In our USENIX CSET and JDMS work my co-authors and I also take a closer look at the experimentation artifacts of malware detection. For instance, we find that current approaches do not consider timing-based calibration of the malware traffic traces prior to using this traffic to salt a background enterprise traffic trace. Thus, timing-based features of the malware traffic may be artificially distinctive, potentially leading to (unrealistically) optimistic classification results.
Science of Security
I am involved in Cyber-Security Collaborative Research Alliance (CSec CRA) with the Army Research Laboratory, Penn State, Carnegie Mellon, UC Riverside, UC Davis, and Indiana University. Our mandate is to develop a new science of security. As part of this effort, I’ve worked on the foundation for representing of operational and environmental knowledge. See our work on operational models. Our goal is to reason about both current and future states of a cyber-operation to make optimal decisions.