Design Methodologies and Tools for Adaptive Machine Learning at the Network Edge

Machine learning technology, such as technology associated with deep learning, is having great impact on numerous sectors of societal importance, such as healthcare, transportation, agriculture, and computer security. Modern applications of machine learning typically involve multiple layers of processing complexity ranging from high-complexity processing on cloud computing servers, where processing resources are abundant, to lower complexity processing on sensor nodes or on mobile devices that are operated by end users. This latter extreme of the processing layers, typically referred to as the network edge, involves critical challenges due to the limited processing resources and stringent energy consumption constraints that are characteristic of edge-based machine learning applications. These challenges involve navigating complex trade-offs among knowledge extraction accuracy from the applied machine learning methods; real-time performance; energy efficiency; and the cost of the devices that are employed for sensing, processing, and communication.

The research program of the CominLabs Chair addresses these challenges through investigation of systematic design methodologies and strategic use of adaptive information processing for edge-based machine learning applications. The proposed research plan will build on existing strengths at IETR Rennes (VAADER Team, INSA), INRIA Rennes, IRISA and Lab-STICC in design tools, embedded computing, formal methods, and image processing to establish new interdisciplinary collaborations that are geared toward enabling novel applications of machine learning through methods for streamlined processing at the network edge.

Comments are closed.