Project1 : Computer Vision

My primary focus in computer vision research lies in the realm of 3D computer vision. Specifically, I am engaged in the areas of Stereo/Monocular depth estimation and 3D reconstruction. 1) In Stereo/Monocular depth estimation, my interest lies in achieving robust depth estimation, especially in challenging scenarios such as: a) Varied lighting conditions like night, snow, rain, etc. b) Dynamic scenes and obstacles encountered in autonomous driving. 2) In the domain of 3D reconstruction, my focus is on leveraging deep learning techniques for the purpose of 3D reconstruction and its application in SLAM (Simultaneous Localization and Mapping).

Project2 : Testing cloud elasticity for Serverless Cloud

Cloud computing is growing in popularity due to rapid provisioning, easy scalability, and pay-as-you-go pricing model. Performance testing on cloud platforms is difficult due to resource contention, cloud providers’ hidden scheduling policies, and passive auto-scaling policies. Additionally, performance testing is more difficult on serverless clouds as the resource abstract level is higher than IaaS clouds’, and auto-scaling stages for serverless clouds are not yet well defined. Our first research goal is to define auto-scaling stages for serverless clouds and develop performance testing methodologies for each auto-scaling stages. The performance uncertainties of serverless cloud are from resource contention during the application execution part and random cold start-up latencies during the environment initiation part. Based on the observations, our second research goal is to develop a new Monte-Carlo-Simulation-based testing methodology to simulate and predict the performance distribution of serverless applications.

Project3 : Testing edge and fog computing

The rapid growth of the Internet of Things (AR, cloud gaming, video streaming, and smart home devices) brings opportunities and challenges to cloud computing industries. Due to the requirement for better performance, local control, and automation, modern IoT has moved away from the old model of processing all requests and data in the centralized cloud. IoT at the edge is becoming increasingly popular thanks to the higher bandwidth and lower latency of edge/fog computing. However, edge/fog computing platforms usually have limited power and computational resources; burst user requests or resource-demanding applications may lead to unacceptable response latencies. And the key to the problem is task offloading. Our research goals are to design testing methodologies to test the effectiveness of existing task-offloading approaches and then develop an AI-based task-offloading approach to better address the issue.

Project4 : Artificial intelligence-based automated cloud testing/recommendation system

Cloud performance fluctuations are mainly from underlying hardware resource contention among cloud neighbors. To obtain accurate performance results, users must test their cloud applications on the actual cloud; the testing cost can be extremely high when a user has multiple applications to be tested. Our research goal is to build an AI-based system that can emulate cloud environments on a local machine to help users accurately obtain performance results at low testing costs.

Project5 : Performance assurance in DevOps

DevOps is a trend towards combining development and operations to shorten the software delivery cycle. To address the lack of systematic techniques to detect performance regression in DevOps, we propose to use the existing development and operational data to assist performance assurance in DevOps. Specifically, the process of detecting performance issues in the context of DevOps is challenging because performance issue detection is usually conducted after the system is built and deployed in the field. Large amounts of resources are required to detect, locate, understand, and fix performance regressions at such a late stage in the software release cycle. State-of-the-art approaches can identify performance issues introduced by code changes. However, they do not consider the impact of performance issues on end-users. Our research goals are to conduct empirical studies on the impact of performance issues on end-users and develop automatic techniques to assist in forecasting performance impacts based on end-user behaviors.