Adaptive Edge Offloading for Image Classification Under Rate Limit
Published:
The paper investigates an edge computing scenario where weak and strong image classifiers located in local devices and an edge server, respectively, collaborate to make the most accurate image classification decisions possible, under the constraint that the number of images that can be offloaded to the strong classifier in the edge server is rate limited using a token bucket mechanism. The paper relies on a reinforcement learning approach to realize a simple policy that maximizes classification accuracy under general image arrival patterns and arbitrary sequences of classification decisions. The code for the system described in the paper is available on GitHub and an extended version of the EMSOFT paper is accessible on arXiv here
Recommended citation: J. Qiu, R. Wang, A. Chakrabarti, R. Guerin, and C. Lu, "Adaptive Edge Offloading for Image Classification Under Rate Limit." IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, Vol. 41, No.11, November 2022. The paper was presented at the ACM International Conference on Embedded Software (EMSOFT), October 2022, Hybrid+Shanghai+Phoenix. https://doi.org/10.1109/TCAD.2022.3197533