My general research interests span edge computing and applied artificial intelligence & data science (especially at the intersection of machine learning and networked & distributed systems), with the objective of making computer systems (e.g., connected vehicles and IoT devices) more Reliable, Scalable, Secure, and Efficient. All my research projects are steered by imperfect real-world datasets with varying quality and inconsistent data frequencies, aiming for solving practical problems by transferring, improving, and delivering trustworthy AI and system solutions from the lab to production, even under unacquainted and adverse environments. My academic advisor is Prof. Specifically, I have accomplished a series of projects based on the Vehicle-EdgeServer-Cloud (VEC) closed-loop framework, and I have further implemented a succession of designs targeting system support for vehicle, EdgeServer, and cloud respectively, contributing to a wide range of edge-enabled applications, including autonomous driving, temporal compressive sensing (e.g., detection on compressed video and reconstruct video if necessary), collaborative learning, connected health (e.g., multimodal signal processing), and large-scale anomaly prediction.
Email  /  LinkedIn  /  Google Scholar  /  ORCID  /  DBLP  /  Web of Science
[1] Integrated Vehicle-EdgeServer-Cloud Framework for Compressive Imaging Processing
Nokia Bell Lab – The CAR Lab Official Collaboration
My role: Project leader
·Introduce the technologies of temporal compressive sensing into the connected vehicle applications. which provides a new and novel direction to achieve faster video processing by executing models on optically compressed video images, i.e., a small number of linear projections of the original video images.
·Experiment results proved that the detection accuracy on the compressed video images (named measurements) is close to the accuracy on reconstructed videos and comparable to the true value, which proves the effectiveness of employing DNN models on the compressed video images. This is the first work that provides a new direction to achieve fast DNN inference.
[2] Reinforcement Learning for Adaptive Video Compressive Sensing
Collaborate with Nokia Bell Lab and Northwestern University
My role: Project leader
·When employing object detection models on the compressed video images, a higher compression ratio (B) is desired to obtain faster inference speed and less transmission costs. However, there is a significant trade-off between B, detection accuracy, and the quality of the reconstructed video.
·One research gap in previous studies is how to adapt the compression ratio under different scenarios. We fill this gap utilizing reinforcement learning (RL) and various DNNs to automatically choose optimal B for different application scenarios. This work takes the technology one step further towards real applications of compressive images.
[3] Implementation and Evaluation of an Edge Computing Framework for Connected Vehicles
Collaborate with Nokia Bell Lab and Northwestern University
My role: Project leader
·The proposed Vehicle-EdgeServer-Cloud (VEC) framework is fully evaluated using our designed roadside platform and outdoor delivery vehicle.
·Our framework has a potential of 18× reduction on the bandwidth requirement, i.e., it can work stably with limited bandwidth of around 320KB/s. This demonstrates its extremely promising applications for connected vehicles (CVs).
[4] Collaborative Learning on the Edges
Research assistant at Wayne State University
My role: Project leader
·Propose a collaborative learning framework on the edges, dubbed CLONE. Two application scenarios for CLONE were explored, including CLONE in the training stage (CLONE_training) and CLONE in the inference stage (CLONE_inference).
·Distributed Electric Vehicle Battery Failure Prediction: As to CLONE_training, we choose the failure of electric vehicle (EV) battery and associated accessories as our case study to show how the collaborative framework can accurately predict failures to ensure sustainable and reliable driving in a collaborative fashion.
·Multi-target Multi-camera Tracking: As to CLONE_inference, we choose multi-target multi-camera tracking as another case study to prove that the collaborative framework can also keep tracking customers based on eight different cameras in a grocery store.
[5] Extensible and Flexible Middleware for Edge Service
Research assistant at Wayne State University
My role: Project leader
·Propose an extensible and flexible middleware to manage the execution of vehicle services, with four key features: i) on-demand model switch, ii) function consolidation and deduplication to eliminate duplicate copies of repeating functions and maximize the reusability of vehicle services, iii) build event-driven applications to reduce workload, and iv) dynamic workflow customization which enables customizing workflow to extend the functionality.
·Our experiment results show that EdgeWare accelerates the execution of services about 2.6× faster compared to the silo approach and save CPU and memory utilization up to around 50% and 17% respectively.
[6] Disk Failure Prediction
Collaborate with Northeastern University
My role: Project leader
·This study covers the disk and server data measured and collected at a large data center. Over, the dataset spans over 64 data center sites, 10,000 server racks and 380,000 hard disks for roughly 70 days. This corresponds to roughly 2.6 million device hours.
·We discovered that performance metrics are good indicators of disk failures and location markers can improve the accuracy of disk failure prediction.
·We also train machine learning models including neural network models to predict disk failures with 0.95 F-measure and 0.95 MCC for 10 days prediction horizon.
[7] Multimodal-based Campus-wide Pandemic Forecasting
Research assistant at Wayne State University
My role: Project leader
·Build a multimodal-based COVID-19 pandemic forecasting platform steering by video, audio, and tweets for Wayne State University to minimize the impact of COVID-19 after resuming academic activities.
·Train diverse models to extract meaningful information from video, audio, and tweets by i) detecting and counting face masks, ii) detecting and counting cough for potential infected cases, and iii) conducting sentiment analysis based on COVID-19 related tweets.
·Fed the multimodal analysis results together with daily confirmed cases data and social distancing metrics into the LSTM model to predict the daily increase rate of confirmed cases for the next seventh day.
[8] Lane Detection/Prediction Enhancement under Challenging Scenarios
Intern project in GM
My role: Project leader
·Focus on the lane detection/prediction enhancement under challenging scenarios with severe occlusion and extreme lighting conditions (e.g., rainy and snowy days) when the perception system of the connected and autonomous vehicle (CAV) loses the lane detection and urgently needs higher-level semantic analysis of lane markings.
[9] A comparison of End-to-End Architectures for Connected Vehicles
Toyota InfoTeach Labs - The CAR Lab Official Collaboration
My role: Project leader
·Examine different types of architectures for vehicle computing, including centralized architecture, decentralized architecture, publish/subscribe architecture, and broadcast architecture.
·Classify essential vehicle applications and discuss their performance requirements (e.g., communication mode, critical latency, data rate per vehicle, and communication range).
·Choose software over-the-air (OTA) update as our case study and propose an edge-based architecture for connected vehicles, named EdgeArC.
[10] Teleoperation Technologies for Enhancing Connected and Autonomous Vehicles
Research Assistant at Wayne State University
My role: Project leader
·Fully autonomous driving is not yet completely feasible. Teleoperation promises to be an important interim solution for bridging the gap between current autonomous driving capabilities and the widespread adoption of autonomous vehicles, where a remote operator can supervise vehicles with manual intervention via wireless communication.
·We explore technical considerations, enhancement techniques, evaluation criteria, and summarize research efforts of automotive participants and teleoperation use cases.
[11] The Emergency of Vehicle Computing
Research Assistant at Wayne State University
My role: Project leader (vision paper)
·With the proliferation of onboard computing and communication capabilities, we envision that connected vehicles will be serving as a mobile computing platform in addition to their conventional transportation role for the next century.
·We also discuss vehicle computing from several aspects, including several case studies, key enabling technologies, a potential business model, a general computing framework, and open challenges.
[12] Cross-Layer Computing Scheduling for Heterogeneous Bare Metal Clusters: A Case Study for Connected and Autonomous Vehicles
Zenlayer – The CAR Lab Official Collaboration
My role: Project leader
·Develop automatic cross-layer computing scheduling approaches for the whole stack from the distributed and heterogeneous infrastructure (i.e., Bare Metal clusters) to the virtual machine/container orchestration platforms (e.g., Kubernetes) and all the way to applications (e.g., games and short videos).
Zenlayer, Sunnyvale, CA, USA April. 2022 – Present
Zenlayer - The CAR Lab Official Collaboration
Mentor: Jim Xu
My role: Develop automatic cross-layer computing scheduling approaches for the whole stack from the distributed and heterogeneous infrastructure (i.e., Bare Metal clusters) to the virtual machine/container orchestration platforms (e.g., Kubernetes) and all the way to applications (e.g., games and short videos).
Toyota Motor North America, Mountain View, CA, USA May. 2021 – Present
Toyota InfoTech Labs - The CAR Lab Official Collaboration
Mentor: Nejib Ammar , Group members: Akila Ganlath and Haoxin Wang
My role: Conduct a comprehensive architecture requirement analysis for connected vehicles (CVs) based on diverse use cases, explore current mainstream architectures for qualitative comparisons, and propose an edge-based architecture for vehicle software over-the-air update, considering the collaboration between CVs, edge servers, and cloud data centers.
Highlight: The first-author collaboration paper has been published in the Fifth International Conference on Connected and Autonomous Driving (MetroCAD’22) [Paper] Toyota proposal has been granted $50K to support this one-year research project (May 2021 - May 2022), and we are continuing our collaboration to implement an end-to-end architecture for CVs.
Ford Motor Company, Detroit, MI, USA May. 2021 – Present
Ford - The CAR Lab Collaboration
Mentor: Oleg Gusikhin , Group members: Omar Makke
My role: Propose an edge-computing advanced analytic platform to perform noise reduction and anomaly detection, explore the impact of different data frequencies on the performance of applications, employ novel unsupervised domain adaptation for automatic data annotation, and leverage temporal compressive sensing to reduce data transmission latency and bandwidth for vehicle fleets.
Highlight: We have submitted an industry-university research program proposal to Ford, which was recommended for funding to support two-year research projects.
General Motors (GM), Detroit, MI, USA Jun. 2020 – Dec. 2020
Research and Development Intern
Vehicle Health Management Group, Propulsion Systems Research Lab
Mentor: Wen-Chiao Lin , Manager: Yilu Zhang , Director: Paul Krajewski
My role: Develop novel approaches supporting enhanced lane detection/prediction for autonomous vehicles, especially under challenging scenarios with severe occlusions or extreme lighting conditions (e.g., night and rain) when the Advanced Driver Assistance System (ADAS) loses the lane detection.
Highlight: GM patent application was filed in 5/10/2022 and application number is 17/740625. Our two intern work reports have been approved and published internally at GM.
Nokia Bell Labs, Murray Hill, New Jersey, USA Oct. 2019 – Dec. 2021
Nokia Bell Labs – The CAR Lab Official Collaboration
Mentor: Xin Yuan
My role: Conduct three projects one by one. Introduce temporal compressive sensing technologies into connected vehicle systems, realize adaptive temporal compressive sensing under diverse driving environments, design, implement and evaluate a proposed vehicle-edge-cloud closed-loop framework.
Highlight: One first-author collaboration paper has been published in the Fifth ACM/IEEE Symposium on Edge Computing (SEC’20) [Paper] [Slides]. The other two papers are under review.
Highlights: Served as the instructor (part-time faculty) and teaching assistant for six semesters in the Department of Computer Science at Wayne State University, including teaching junior/senior undergraduate courses and research-oriented graduate courses. The faculty individual report denoted the mean and median ratings of my Student Evaluation of Teaching (SET) are 4.4/5 and 5/5, respectively [ ].
Instructor (Part-time Faculty), CSC 4920 Introduction to Computer Networks ;Fall 2022
Department of Computer Science, Wayne State University
· Preparing lecture materials, initiating discussions, holding office hours, and designing and grading in-class quizzes, assignments, and exams for 15 undergraduate students.
Teaching Assistant, CSC 5250 Network Distributed and Concurrent Programming Fall 2018
Department of Computer Science, Wayne State University
· Elaborating on basic principles, moderating course presentation sessions, and grading course assessments for around 10 graduate students.
Teaching Assistant, CSC 4996 Senior Project and Computer Ethics Winter 2018
Department of Computer Science, Wayne State University
· Guiding three teams of 12 undergraduate students to complete their graduation capstone projects.
· Coaching teams on establishing appropriate project milestones, holding weekly meetings, and providing technical instruction to students.
· Managing projects including CarPool iOS Application, Directed Graph Viewer (Paper Trail Web Application), and GM Big Data Analytics.
Instructor, CSC 1101 Problem Solving and Programming Laboratory Fall 2016, Winter 2017, Fall 2017
Department of Computer Science, Wayne State University
· Preparing lectures slides, holding office hours, and designing and grading hands-on exercises and assignments for around 20 undergraduate students to understand programming basics with C++.
· Student Evaluation of Teaching (SET): a mean score of 4.5/5 and a median score of 5/5 [ ].
Highlights: Published and accepted 14 (10 first-author) peer-reviewed research articles in premium CS conferences and journals including FAST, SEC, ICDCS, WACV, IEEE Internet of Things Journal, and IEEE Internet Computing. Two first-author papers have been submitted and are currently under review.
My role: penned and polished one chapter related to the teleoperation technologies for autonomous vehicles.
Disk Failure Prediction Tookit [Link]
This repository contains the code for the paper “Making Disk Failure Predictions SMARTer!”. It includes the source codes of five machine learning models which are trained for disk failure prediction, including the naive Bayes classifier (Bayes), random forest (RF), gradient boosted decision tree (GBDT), long short-term memory networks (LSTM), and convolutional neural network long short-term memory (CNN-LSTM).
Compressive Imaging Processing Tookit [Link]
The source code of the paper “EdgeCompression: An Integrated Framework for Compressive Imaging Processing on CAVs”. It includes different trained models for video compression, vehicle detection, and video reconstruction.
Middleware for Connected Vehicle Service [Link ]
This repository contains the code for the paper “EdgeWare: Towards Extensible and Flexible Middleware for Connected Vehicle Service”, which has four key features: i) on-demand model switch, ii) function consolidation and deduplication to maximize the reusability of vehicle services, iii) build event-driven applications to reduce workload, and iv) dynamic workflow customization.