The site’s research projects will focus on equipping students with skills and knowledge related to big data analytics in cyber-physical systems (CPS). Research projects will involve combinations of Application Areas and Analytics Dimensions shown in the following figure.
Example Projects for Systems and Architecture (D1)
Project 1 (Ongoing) Title: Use of GPU Platforms for Constraint Optimization Architectures
Advisors: Drs. Enrico Pontelli, Son Cao Tran
Description: Distributed constraint-based models have been widely used to model Cyber-Physical Systems (CPS) and support distributed problem solving (e.g., as Distributed Constraint Optimization Problems). Constraint-based optimization algorithms face the challenge of scalability when applied to complex problems; nevertheless, recent research has highlighted the potential of using hard constraints to effectively prune the search space. Propagation of hard constraints, especially global constraints (i.e., hard constraints that affect a large number of problem variables), is often realized using ad-hoc algorithms, frequently based on dynamic programming techniques. The structure exploited by these approaches in constructing solutions makes it suitable to exploit the type of parallelism provided by modern Graphic Processing Units (GPUs)—massively parallel platforms that are based on the Single Instruction Multiple Thread paradigm. Building on our preliminary work, this project explores the use of GPUs to parallelize propagation of hard constraints in both centralized and distributed constraint optimization frameworks.
Research Objectives: Learn propagation algorithms of hard and global constraints, design and implement GPU-based versions of the algorithms, and assess performance.
Student Learning: Students will learn programming GPUs, dynamic programming algorithms for constraint propagation, and gain hands-on experience in parallelizing algorithms and measuring parallel performance.
Project 2 (Ongoing) Title: Application Porting and Benchmarking for Memory Side Processing on EMU Systems
Advisors: Drs. Hameed Badawy, Olga Lavrova
Description: Large amount of CPS data needs to be analyzed using machine learning and data mining techniques (D2). Most existing computing paradigm moves the data to the computing threads. In this project, we will investigate a new paradigm of computing which involves moving the computation threads to the data. We will explore a Field Programmable Gate Array (FPGA) implementation for the memory-side processing technology, which has been introduced by EMU technologies. Dr. Badawy maintains a collaboration with the DOD Laboratory of Physical Sciences (LPS) to explore the full impact of such a paradigm shift proposed by EMU technologies and their migratory memory-side processing paradigm. EMU machines are expected to work well for graph applications and sparse matrix-vector and matrix-matrix applications. The major challenge derives from the need to determine kernels for the prospective applications that can be ported to the Cilk programming language (supported by EMU machines) and optimize them for memory-side processing. One REU student (Matthew Bredin) working on this project was able to generate interesting results.
Research Objectives: Identify application kernels of interest, transform them into optimized Cilk kernels that run efficiently on EMU machines, and benchmark them against regular GPU and x86 implementations.
Student Learning: Students will learn performance profiling, porting, benchmarking, and modeling techniques in applications, systems, and architectures.
Project 3 (New) Title: Exploring Scratch-Pad Memories in Conjunction or In-lieu of Caches in Modern CPUs
Advisors: Drs. Hameed Badawy, Satyajayant Misra
Description: The execution time of CPS tasks is critical to its correct functioning. For example, the monitoring and protection of CPS often need real-time data analysis. To improve efficiency of CPS analysis, fast access storage media (e.g., caches) are often used because caches has been the nearest memory component of modern CPUs. In addition, Scratch-pad memories are also used in embedded systems [5, 84], which are often implemented in CPS. This project will explore the usage of Caches and Scratch-Pad memories in a novel method which bridges the compiler and hardware to manage such storage systems near the CPU. Memory side processing systems, as implemented in the EMU technology machines, which employ no caching mechanism at all and rely on DRAM memory systems, are prime candidates for such an innovative approach. Benchmarking and implementing such systems will happen in the simulation infrastructure of the EMU systems as well as in architecture simulators (e.g., Gem5) and compiler frameworks (e.g., LLVM ).
Research Objectives: Implement a scratch-pad design in a cycle-accurate simulation framework such as Gem5 or EMU with compiler support using the LLVM compiler infrastructure, and test the design.
Student Learning: Students will modify cycle-accurate simulators to implement hardware structures such as Scratch-pad memories and integrate them into application development using LLVM compiler infrastructure.
Example Projects for Models and Algorithms (D2)
Project 4 (Ongoing) Title: Software Agents for Coordinating Devices in Smart Homes
Advisors: Enrico Pontelli, Son Cao Tran
Description: Smart devices (e.g., smart thermostats, smart TVs, smart refrigerators) as well as renewable resources (e.g., solar panels) are becoming increasingly ubiquitous in our homes. The goal of this project is to develop and implement agent-based coordination mechanisms to coordinate and control these devices in a neighborhood of homes aimed at reducing overall energy consumption and minimizing our carbon footprint. A main challenge for these mechanisms is that privacy concerns require that the coordination happens in a distributed (localized) manner, yet result in efficiently and effectively solutions.
Research objective: Model the coordination problem using multi-agent coordination models (e.g., negotiations, distributed constraint optimization, decentralized Markov decision processes) and tailor the algorithms to specific application domains.
Student Learning: Students will learn the fundamental theories of power systems and multi-agent coordination systems. They will also get the experience of implementing these mechanisms on Raspberry Pis, which they will deploy on an actual microgrid.
Project 5 (New) Title: Dimensionality Reduction for Uncertain Data
Advisors: Drs. Tuan Le, Huiping Cao, Satish Ranade
Description: Real-world smart grid data can be uncertain. The uncertainties may arise from many sources such as measurement errors caused by sensor inaccuracies, missing data and noises due to equipment malfunction or cyber-attack. With uncertainty, data points may have features whose values are missing or have features with multiple values. This raises many challenges for applying machine learning techniques such as dimensionality reduction for mining smart grid data. Dimensionality reduction (DR) is widely used in several smart grid applications, for instance, demand side management, detecting household characteristics based on smart meter data, power system event classification, phase identification in smart grid, and event detection. In these applications, DR methods are mainly used to extract key features from data for achieving computational efficiency as well as improving the analysis accuracy. However, most of the traditional DR methods (e.g., PCA or t-SNE) are not designed to deal with data uncertainties, and this may affect the performance of those smart grid applications. In this project, we aim to develop non-linear dimensionality reduction methods for uncertain data that can handle data points whose features may be missing or have multiple values. These challenges have been recently considered in the context of classification and visualizing streaming high dimensional data. However, the authors employ incremental PCA which is linear and therefore may not substantially capture the inherent non-linear structures caused by the complexity of the smart grid.
Research Objective: Develop non-linear dimensionality reduction methods for uncertain data.
Student Learning: Students will learn state-of-the-art methods for dimensionality reduction.
Project 6 (New) Title: Exploring Privacy and Security Issues in Smart Grid Networks
Advisors: Drs. Roopa Vishwanathan, Huiping Cao, Satyajayant Misra
Description: The smart grid contains a wealth of information including customers’ energy consumption patterns and preferred energy providers. There are natural privacy concerns about customers’ identity and usage data being made available to parties without customers’ consent. One way to deal with this is to borrow techniques from cryptography and privacy preserving data mining such as private information retrieval and oblivious transfer of data. Another exciting direction to explore is to see if relatively new advances in cryptography, such as attribute-based cryptosystems , and/or predicate encryption can be applied to the smart grid environment. These protocols and techniques provide a very high degree of security, but they can be computationally expensive to implement. The main challenge is to design scaled-down variants of these well-known protocols to suit the resource-constrained smart grid environment. If successful, this would result not only in robust implementations, but also in exciting new definitions and theoretical frameworks that are custom-designed for cryptography in a smart grid environment.
Research Objective: Identify the privacy and security problems in the smart grid setting, and explore using cryptographic protocols to solve them, in an efficient way.
Student Learning: Students will learn the fundamentals of cryptography, and applying them in a practical setting. Students will also be exposed to basic crypto theory and proof techniques.
Project 7 (New) Title: Studying Privacy, Security in Blockchain Enabled CPS Applications
Advisors: Drs. Roopa Vishwanathan, Satyajayant Misra, Olga Lavrova
Description: Blockchain technology has been deployed in a variety of applications, ranging from financial networks, healthcare services to IoT networks. In this project we plan to study the design of blockchain-enabled CPS applications using cryptographic protocols and primitives. We plan to focus on smart grid services and their related privacy, security issues. We have designed secure and private ways to conduct transactions in distributed credit networks (DCNs). The rising popularity of DCNs stem from their capability to enable direct exchanges between users in exchange for users accepting counter-party credit risks. There are several exciting open problems in this area, and our recent previous works have just scratched the surface. For example, our work did not consider the challenging problem of rebalancing in credit networks, which addresses the question of what happens when a user’s credit links get exhausted. In this case, does she still remain part of the network? How do we replenish the credit links? Another issue that needs attention is a real-world permissioned blockchain implementation of our work, and its associated challenges. We believe both these problems are very suitable for undergraduate students.
Research Objective: Study privacy and security issues in blockchain enabled CPS applications
Student Learning: Students will learn how to implement blockchain enabled applications, and will work on well-known platforms such as Hyperledger, and will learn how to write Solidity code.
Project 8 (New) Title: Vision-based Multiple Target Tracking
Advisors: Drs. Liang Sun, Parth Nagarkar
Description: An unmanned aerial vehicle (UAV) is able to carry a gimbaled system that consists of multiple visual sensors (e.g., RGB, IR, in-depth cameras, etc.). Due to the limited field of view of visual sensors, complicated tasks demand multiple UAVs to work together in a cooperative manner. This research project will have direct applicability to topics such as intelligence, surveillance, and reconnaissance (ISR), border security, law enforcement, search and rescue, wildfire tracking, and farming/ranching land utilization. In these applications, it is desired that a small number of cooperative UAVs are able to track a large number of mobile ground targets [30–32]. Target extraction algorithms associated with sensor management/fusion techniques are needed to quickly identify the targets of interest from a series of image frames [41, 45, 86]. The camera orientation will continuously be changing due to the need to observe a specific area in realtime. Slew rate of the gimbal and the time period that the camera system needs to remain focused are vital factors. These challenges call for fast and effective image processing algorithms.
Research Objective: Investigate the integration of cooperative UAV techniques with image processing and machine learning to analyze aerial images for target tracking.
Student Learning: Students will learn techniques of target tracking and machine learning.
Example Projects for Visualization (D3)
Project 9 (On-going) Title: Augmented Reality Home Interfaces for Smart Energy Use
Advisors: Drs. Zachary O. Toups, Son Cao Tran, Parth Nagarkar
Description: Smart appliances and networked power sensors are starting to surface power usage data to users, empowering them to make smart decisions about energy use. The proposed research leverages these technologies and connects them with augmented reality to superimpose a visualization on the real environment, to enable users to explore their energy usage. The proposed system will surface energy use information, both in real time and over extended periods, with creative visualizations embedded in the real environment. We hypothesize that these real world embedded visualizations will impact energy use behaviors. In this project, we will begin with a stationary virtual reality experience, using existing hardware – a low-fidelity prototype of the proposed work. Later we will build an augmented reality system for users, which will enable them to explore an indoor environment and obtain real-time information about energy usage. Later phases of the research will develop more involved, persuasive experiences (such as games) to support long-term behavior change around energy usage.
Research Objective: Determine (i) best practices around presenting environmental information through real-world interfaces and (ii) best means of presenting information to engage users in long-term behavior change around energy usage.
Student Learning: Students will learn the fundamental theories of interface design and evaluation, and work with advanced technology, such as augmented and virtual reality equipment and wearable computers.