Boyuan Guan is a Ph.D. candidate in the Knight Foundation School of Computing and Information Sciences (KFSCIS) at Florida International University (FIU), who works with Dr. Liting Hu in the area of system optimization for Big Data applications. Boyuan is also the lead developer of the FIU GIS Center. Boyuan has led the design and research of the intelligent recommendation library system dpSmart and the paper “dpSmart: A Flexible Group Based Recommendation Framework for Digital Repository Systems” in the IEEE BIG DATA CONFERENCE in 2019. Boyuan is also involved in several studies on optimization of CaaS systems and Big Data streaming, such as “Exploiting the Spam Correlations in Scalable Online Social Spam Detection” and “Towards Adaptive Replication for Hot/Cold Blocks in HDFS using MemCached”. Boyuan is currently leading the system design and architectural work for the Security and Research Hub project, funded by U.S. Southern Command.
Over the past decade, the popularity of machine learning applications such as recommender systems, image recognition, real-time alerts, event detection, natural language processing, and online streaming analytics has increased dramatically. In both the commercial and scientific sectors, rapid environment setup and application deployment is a compelling requirement. As a result, more and more organizations are opting for cloud environments when deploying machine learning applications, rather than setting up the environment from scratch themselves. Cloud computing resources such as server engines, orchestration, and underlying server resources are provided to users as a service by a cloud provider. However, due to the scale and complexity of machine learning, it is difficult to efficiently provision and manage data and resources even in the cloud computing environment. This hinders and reduces productivity. Serverless computing as a stateless cloud computing model is an emerging solution that offers significant efficiency and cost benefits for event-driven applications in the cloud environment, including artificial intelligence (AI) and machine learning applications. With serverless computing, the complexity of the machine learning system is minimized, and it is flexible and easy to manage. On the other hand, the inputs of machine learning applications are usually in a real-time or near real-time data stream format. Operating and managing serverless machine learning services in clouds encounters many limitations, such as latency and privacy. Local, distributed edge computing nodes that are not accessible to users can solve these challenges of serverless cloud AI applications.
In this dissertation, we present a method that consists of three main components and provides a real approach for making classical machine learning applications compatible with the serverless edge environment. The three components are presented below: (1) the first component is a practical and flexible framework to transform the machine learning application into a Function-as-a-Service (FaaS) design (2) the second component is a general framework to leverage the edge devices and build the data pipeline for communication between the cloud and the edge environment, (3) the third component discusses and evaluates the existing serverless edge framework and presents the optimized resource-aware approach specifically for the machine learning application.