Robert J. Walls147 Fuller Labs
Worcester, MA 01609
The Cake Lab Google Scholar
Students!: I am currently looking for motivated and qualified students. If you are a WPI student, please send me an email or stop by. If you are outside of WPI, I invite you to apply and then contact me. Use keyword fignewtons for a more rapid response.
I am an assistant professor in the Department of Computer Science at Worcester Polytechnic Institute and a proud member of The Cake Lab group. My current interests focus on systems security and performance and my projects often lie at the intersection of software and hardware.
Previously, I was a postdoctoral scholar in Department of Electrical Engineering and Computer Science at The Pennsylvania State University working with Prof. Patrick McDaniel. Before that, I attended the School of Computer Science at the University of Massachusetts advised by Prof. Brian Levine.
I’ve had the opportunity to work on a number of interesting research projects during my career. Checkout The Cake Lab’s website for the most up to date information about my work.
Embedded Systems Security
Embedded systems form the core of critical infrastructure, perform auxiliary processing on mobile phones, and permeate homes as smart devices. Yet, embedded software security lags behind traditional desktop security. While myriad defenses exist for general-purpose systems (e.g., desktops and servers), embedded systems present several unique challenges for software security such as greater hardware diversity, limited resources (e.g. memory and power), and lack of support for common abstractions like virtual memory. Our work in this area includes defenses for protecting embedded software and RTOS kernels from memory errors, such as Kage (USENIX’22), Silhouette (USENIX’20), and Recfish (ECRTS’19).
Secure Deep Learning
ML models are valuable intellectual property due to the investment and expertise required to gather training data and construct the model. To monetize these models, companies often make them available as a service via APIs over the web. Further, the model owners often rely on others’ hardware, such as cloud providers or end-users, for model execution. Our CVPR’21 work demonstrates the feasibility of data-free model extraction attacks, i.e., a type of attack that does not require knowledge of the underlying training dataset. In our IC2E’21 study, we identify performance bottlenecks that stymie current attempts to run models in trusted execution environments.
The RIPCORD project proposes a new infrastructure for improving the performance of deep learning model serving. In our Performance’20 and Performance’21 studies, we considered the performance of current GPU concurrency mechanisms. We examined scheduling at the microarchitectural level and found that the lack of fine-grained preemption mechanisms, robust task prioritization options, and contention-aware thread block placement policies limits the effectiveness of existing mechanisms.
Below are some of the previous projects I have had the privilege to work on.
Web Security and Privacy
Domain names have become the Internet’s de facto root of trust. In practice, they are also a root of insecurity as common security systems depend on the unfounded assumption that domain ownership remains constant; this leaves users vulnerable to exploitation when domain ownership changes. In our [Oakland 2016][oakland16] we find that many seemingly disparate security problems share a root cause in residual domain trust abuse.
Online advertising is one of those little annoyances that we all have to deal with. “Not so!” Said the plethora of ad blocking extensions promising to improve your browser experience. Not only do they block ads, they also claim to help preserve your privacy and protect you against the growing trend of malicious advertisements. In our IMC 2015 work my co-authors and I take a closer look at the most popular ad blocking software. We find that ad blockers are not quite what they appear to be.
Mobile phones contain evidence that is invaluable for criminal investigations. However, commercially-available forensic tools must be hand-tailored to each phone model. If no tools support the target phone, then extracting the phone’s information requires investigators to examine the stored data byte by byte. To address this problem, I’ve developed general algorithms and techniques for recovering information from phones even if the exact storage format is unknown or the data has been logically deleted.
DECODE is an inference engine that extracts meaningful information from raw byte streams. Read more about it here. Liftr incorporates investigator feedback and relevance graphs to improve the results of inference engines like DEC0DE. Paper here.
Science of Security
Perhaps the most ambitious projects I’ve been involved with is the 10-year Cyber-Security Collaborative Research Alliance with the Army Research Laboratory, Penn State, Carnegie Mellon, UC Riverside, UC Davis, and Indiana University. The project’s mandate is to develop a new science of security. As part of this effort, I’ve worked on the foundation for representing operational and environmental knowledge—see my work on ontologies here—with the goal of reasoning about both current and future states to make optimal security decisions.
Below is a partial list of my recent publications.