Incremental Online Learning Algorithms Comparison for Gesture and Visual Smart Sensors

Abstract

Tiny machine learning (TinyML) in IoT systems exploits MCUs as edge devices for data processing. However, traditional TinyML methods can only perform inference, limited to static environments or classes. Real case scenarios usually work in dynamic environments, thus drifting the context where the original neural model is no more suitable. For this reason, pre-trained models reduce accuracy and reliability during their lifetime because the data recorded slowly becomes obsolete or new patterns appear. Continual learning strategies maintain the model up to date, with runtime fine-tuning of the parameters. This paper compares four state-of-the-art algorithms in two real applications: i) gesture recognition based on accelerometer data and ii) image classification. Our results confirm these systems' reliability and the feasibility of deploying them in tiny-memory MCUs, with a drop in the accuracy of a few percentage points with respect to the original models for unconstrained computing platforms.

A Machine Learning-Oriented Survey on Tiny Machine Learning

“…Another effective approach is to use dynamic architectures that can adapt their structure to accommodate new tasks. For instance, in [149], a regularization-based approach for an IoT scenario is presented, in which MCUs are exploited as edge devices for data processing considering two tasks: gesture recognition based on accelerometer data and image classification.…”

Deep learning and transfer learning for device-free human activity recognition: A survey

“…Up till now, universal frameworks enabling full-network fine-tuning/training on MCUs [76]- [78] are still scarce, while they are mostly dedicated to vision tasks based on 2D-Convolutional Neural Network (CNN) architecture. Compared with fine-tuning the whole network, on-device transfer learning (ODTL) [26], [80], [81] that selectively updates only specific layers, e.g., the last dense layer, offers a preferable solution that better balances between resourcefriendliness and performance gains. For example, [49] reports an average performance loss of only 1.5% by shifting from full-network training to partial transfer learning (only dense layers) in the personalization task on the WISDM data set, while 70.4% latency and 36.2% memory footprint can be saved during the learning process.…”