Schedule and Details

More Details

Sept 1, Fri, 12:00 PM at EB2 3211 (CPS Lunch)

Title: A New Paradigm of Accelerating Privacy Preserving Machine Learning as A Service
Speaker: Wujie Wen
Abstract: Fueled by data proliferation and machine learning advancement, intelligence is becoming a household brand from cloud to edge, transforming every walk of life. While enticing, current smart data services, especially machine learning as a service (MLaaS) on cloud, face ever-increasing privacy concerns. For example, protecting the confidentiality of clients’ data is crucial for sensitive applications such as healthcare and financial analysis. Unfortunately, achieving privacy comes at the cost of prohibitively high inference latency or significantly downgraded accuracy. In this talk, I will discuss our recent progress on achieving and accelerating privacy-preserving machine learning inference. The talk starts with “CryptoGCN”-the very first endeavor to facilitate and accelerate Homomorphic Encryption (HE)-based private graph convolutional neural network (GCN) inference, followed by “SpENCNN”-a HE-based fast convolution neural network (CNN) inference framework built upon the co-design of encryption operation-aware model sparsity and the single-instruction-multiple-data (SIMD)-friendly data packing. I will also share our key finding from these studies, which is to significantly decrease the memory and computation footprint needed in costly encryption operations by orchestrating ciphertext encoding, sparsity pattern and model architecture design based on an individual network’s unique computing pattern. The prospects on the research along this direction will also be given at the end of this talk.

Sept 8, Fri, 12:00 PM at EB2 3211 (CPS Lunch)

Title: Towards Precision Sensing in IoT for Human Machine Interface
Speaker: Chenhan Xu
Abstract: Nowadays, IoT devices are encroaching on every environment of our lives, including but not limited to homes, vehicles, and offices. However, due to the complexity of human activities, unavoidable variance in signal scaling, and privacy-preserving requirements, only limited human information can be cognized by the IoT environment, leading to a great gap between IoT intelligence and human demands. In this talk, I will introduce how we build sensing and computing systems that aim for precise, secure, intelligent, and broad-spectrum human-computer interaction (HCI). In particular, I will present our work that is the first to exploit high-precision millimeter wave direct sensing of the human vocal system to solve acoustic noise problems on conventional voice sensing for decades, which is a fundamental refactor of voice sensing and enables new voice computing applications. I will also present our work on a next-generation voice-user interface that enables pervasive heart-sensing functions. I will conclude my talk with future directions in IoT systems for humans.

Sept 22, Fri, 12:00 PM11:30 PM at EB2 3211 (CPS Lunch)

Title: Formal Synthesis-Driven Noninvasive Attack-Resiliency Analytics for Internet of Things-Based Smart Systems
Speaker: Mohammad Ashiqur Rahman
Abstract: As cyber-physical systems (CPSs) evolved with the increasing availability of the internet of things (IoT) and improved communication infrastructures, their security has been a major concern to both practitioners and academicians. Empirical analysis or systematic verification of anticipated attacks, often considering control functions and attack properties in isolation, are not suitable for adequately realizing the threat space or making cost-effective hardening. Formal reasoning-based artificial intelligence has been proven advantageous for threat analysis because they are provable and noninvasive and have the power to model the system holistically. Furthermore, ML-based control techniques are increasingly used in modern IoT/CPSs, which often introduce different threat characteristics than typical physics-based controllers. These ML-based control functions lack systematic mathematical analyses. Therefore, we require mechanisms to conceptualize the control logic from the ML models to facilitate formal threat synthesis concerning such controllers. This talk will present several of our research works where we develop formal synthesis-based analytics to find potential attacks against smart systems, where ML is often used to increase the controller’s performance or add security to it.

Short Bio: Dr. Mohammad Ashiqur Rahman is an Associate Professor in the Department of Electrical and Computer Engineering and the School of Computing and Information Sciences at Florida International University. He obtained a PhD in computing and information systems from the University of North Carolina at Charlotte (UNC Charlotte) in 2015. Previously, he received BS and MS in computer science and engineering from Bangladesh University of Engineering and Technology (BUET). Dr. Rahman’s primary research interests cover a wide area of computer networks and cyber-physical systems (CPS). His research focus primarily includes computer and information security, risk analysis and security hardening, secure and dependable resource allocation, and distributed computing. His research is primarily funded by NSF, DOE, and DOD. He is currently leading multiple grants on CPS
security. Dr. Rahman coauthored a book and several book chapters and published over 100 peer-reviewed journal and conference papers. He served on the organization and technical program committees (TPCs) for various IEEE and ACM conferences. He served as the TPC Co-Chair of IEEE/IFIP NOMS 2023.

Sept 29, Fri, 12:00 PM at EB2 3211 

Title: Detecting Code Execution Vulnerabilities in Cloud Server Systems
Speaker: Fogo Tunde-Onadele
Abstract: Cloud systems are a popular platform for deploying production software. Accordingly, security attacks to cloud server systems can have extensive real-world impact. Although existing intrusion detection systems find anomalies in system behavior, they do not inform developers about the underlying code defects. Thus, developers spend resources to analyze and fix the security bugs. In this work, we leverage studied code patterns to proactively protect cloud server systems from a dominant category of security bugs that are due to the improper execution restrictions. We present work towards a new static analysis approach, ExeScope, that combines call graph analysis with data-flow analysis to efficiently detect code execution bugs. We evaluate our scheme over six common vulnerability exposures (CVEs) with unique root causes in five Java cloud server systems. The preliminary results show that ExeScope can detect the vulnerable function root causes but still suffers from false positives. We compare ExeScope to existing Java bug finding tools and discuss next steps for addressing the false positive issue.

Oct 6, Fri, 12:00 PM at EB2 3211 (CPS Lunch)

Title: ROS2 Scheduling with Dynamic Priorities
Speaker: Kurt Wilson
Abstract: Robot Operating System (ROS) is the most popular framework for developing robotics software. Typically, robotics software is safety-critical and employed in real-time systems requiring timing guarantees. ROS2 attempts to address the needs of real-time systems, but lacks proper scheduling guarantees, which severely affects the response time of ROS2 applications. We propose a deadline-based scheduling strategy for the ROS2 executor, and modify it to include both deadline-based and priority scheduling. We also analyse and address potential edge cases in multithreaded ROS2 systems. Our modifications increase the predictability of ROS2, and allow for the development of real-time systems. Our modifications can be dropped into existing ROS2 systems with minimal changes.

Oct 19, Thu, 3:30 PM at EB2 3211 

Title: Pinpointing resource wastage with profiling
Speaker: Pengfei Su
Abstract: Software packages have become increasingly complex with millions of lines of code, sophisticated control and data flow, and escalating levels of abstraction. This complexity often introduces resource wastage across software stacks, making it practically impossible for users to pinpoint them. Profilers abound in the community to aid software developers in understanding program behavior. However, classic profiling techniques focus on execution hotspots with a myopic view of resource wastage; for example, they can hardly diagnose whether a resource is being used in a productive manner that contributes to the overall efficiency of a program. In this talk, I will describe various forms of resource wastage, which pervasively exist in modern software packages and expose great potential for optimization. I will discuss the design of an instrumentation-based profiler that identifies wasteful memory operations, which guides nontrivial performance improvement. Furthermore, I will discuss the design of a lightweight sampling-based profiler that pinpoints execution anomalies, an overlooked indicator of resource wastage. Last but not least, I will show our recent efforts in saving scarce GPU memory resources.
Short Bio: Pengfei Su is an assistant professor in the Department of Computer Science and Engineering at the University of California, Merced. He obtained his Ph.D. from William & Mary in 2020. His research interests lie in high-performance computing, programming languages, and program analysis, with a focus on providing tools support for analyzing and optimizing software inefficiencies. He has published papers in peer-reviewed conferences and journals, including ICSE, ESEC/FSE,  SC, PPoPP, CGO, ASPLOS, and TACO. His papers received Best Paper Award at PPoPP'19 and Distinguished Paper Award at ICSE'19.

Oct 20, Fri, 12:00 PM at EB2 3211 

Title: Accelerating HPC Applications Using Neural Network-based Surrogates
Speaker: Wenqian Dong
Abstract: Neural network-based surrogate, aiming to use neural networks to accelerate time-consuming computation in scientific applications, is promising in high-performance computing (HPC). However, there is a lack of tools that can help domain scientists automatically apply neural network-based surrogate models to HPC applications. In this talk, we introduce a framework, named Auto-HPCnet, to democratize the usage of neural network-based surrogates. Auto-HPCnet is the first end-to-end framework that makes past proposals for the neural network-based surrogate model practical and disciplined. Auto-HPCnet introduces a workflow to address unique challenges when applying the approximation, such as feature acquisition and meeting the application-specific constraint on the quality of the final computation outcome.  We show that Auto-HPCnet can leverage neural networks for a set of HPC applications and achieve 5.50 x speedup on average (up to 16.8 x speedup and with data preparation cost included) while meeting the application-specific constraint on the final computation quality.
Short Bio: Dr. Wenqian Dong is an assistant professor in the KFSCIS department at Florida International University (FIU). She earned her Ph.D. in EECS at the University of California, Merced, in Spring 2022. Recently, she is selected for the IEEE-CS Technical Community on High-Performance Computing (TCHPC) Early Career Researchers Award for Excellence in High-Performance Computing. Her research focuses on three main areas. She has contributed significantly to scientific machine learning, particularly in using machine learning to speed up HPC applications. Her work is showcased in conferences like SC’19 and SC’20. Wenqian has excelled in automatic machine learning, concentrating on creating machine learning models for HPC applications. Her papers in VLDB’21, HPDC’23, and ASPLOS’22 highlight her notable contributions. She’s skilled in optimizing system performance, aiming to enhance HPC applications’ quality and efficiency through system optimization. Her work presented at conferences like ICS’21, Eurosys’21, ICPP’18, and Parallel Computing’23 illustrates her dedication to this field. Her work has generated real impacts in the HPC community. For example, her work on power grid simulation using ML led to 3.28 times performance improvement and highlighted Newswise as a DOE science innovation. Furthermore, Wenqian is committed to enriching the HPC community. Her commitment is apparent in her various roles as an organizer for the MLBench’23 workshop, and as a member of the Technical Program Committee (TPC) for the IEEE Cloud 2023, IEEE Cluster 2023, AI4Science 2022 workshop, and the GPGPU 2023 workshop.