REU Supplement to CAREER: Integrated and End-To-End Machine Learning Pipeline For Edge-Enabled IoT Systems: a Resourceaware and QoS-Aware Perspective

Grants and Contracts Details

Description

Abstract The recently awarded grant proposal, entitled “CAREER: Integrated and end-to-end machine learning pipeline for edge-enabled IoT systems: a resource-aware and QoS-aware perspective” (Federal Award ID Number: 2340075), proposed an integrated resource-aware and QoS-aware data pre-processing and model training system that can automatically optimize the system’s performance by dynamically balancing the allocation of resources between data pre-processing and model training. Although data pre-processing and model training have been separately studied in the literature, none of the previous work consider a wholistic and integrated system design for such problem. Combining data pre-processing with model training enables adaptable preprocessing techniques tailored to specific data characteristics and model needs. This approach accommodates various sensor data conditions and optimizes data representation for precise predictions. Unlike prior studies, we optimize both data pre-processing (including cleaning and reduction) and model training concurrently, mindful of network constraints. This integration of decision-making processes creates a feedback loop, enhancing both facets and resulting in improved data representation alignment and system performance in many resources constrained applications such as Internet-of-Things (IoT) applications. This research tackles challenges in distributed data analytics, sparse resource management, handling unlabeled and noisy data and addressing device failures. Techniques such as network coding, coded computing, reinforcement learning, compression, and resource management are employed to mitigate these challenges effectively.
StatusActive
Effective start/end date3/1/242/28/29

Funding

  • National Science Foundation

Fingerprint

Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.