Self-Learning Pipeline for Low-Energy Resource-Constrained Devices

The trend of bringing machine learning (ML) to the Internet Things (IoT) field devices is becoming ever more relevant, also reducing overall energy need applications. ML models are usually trained in cloud and then deployed on edge devices. Most IoT generate large amounts unlabeled data, which expensive challenging annotate. This paper introduces self-learning autonomous inferencing pipeline (AEP), deployable a resource-constrained embedded system, can be used for unsupervised local training classification. AEP uses two complementary approaches: pseudo-label generation with confidence measure using k-means clustering periodic one supported classifiers, namely decision tree (DT) k-nearest neighbor (k-NN), exploiting pseudo-labels. We tested proposed system datasets. AEP, running STM NUCLEO-H743ZI2 microcontroller, achieves comparable accuracy levels as same-type actual labels. makes an in-depth performance analysis particularly addressing limited memory footprint support remote robustness.