TinierHAR: Towards Ultra-Lightweight Deep Learning Models for Efficient Human Activity Recognition on Edge Devices

2025-07-10

AI Summary

To address the trade-off between accuracy and efficiency in human activity recognition (HAR) on resource-constrained wearable devices, this paper proposes an ultra-lightweight end-to-end spatiotemporal feature extraction network. Methodologically, it integrates residual depthwise separable convolutions, gated recurrent units, and a novel temporal aggregation mechanism. We conduct the first systematic ablation study of spatiotemporal components to quantify module contributions and derive design principles for edge-oriented HAR models. Evaluated across 14 public datasets, our model reduces parameter count by 2.7× and MACs by 6.4× compared to TinyHAR, while maintaining competitive average F1-score—achieving the state-of-the-art efficiency–accuracy balance. All code is fully open-sourced to facilitate reproducibility and standardization of edge-intelligent HAR.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Computer Vision: Video Understanding & Activity Analysis

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Human Activity Recognition (HAR) on resource-constrained wearable devices demands inference models that harmonize accuracy with computational efficiency. This paper introduces TinierHAR, an ultra-lightweight deep learning architecture that synergizes residual depthwise separable convolutions, gated recurrent units (GRUs), and temporal aggregation to achieve SOTA efficiency without compromising performance. Evaluated across 14 public HAR datasets, TinierHAR reduces Parameters by 2.7x (vs. TinyHAR) and 43.3x (vs. DeepConvLSTM), and MACs by 6.4x and 58.6x, respectively, while maintaining the averaged F1-scores. Beyond quantitative gains, this work provides the first systematic ablation study dissecting the contributions of spatial-temporal components across proposed TinierHAR, prior SOTA TinyHAR, and the classical DeepConvLSTM, offering actionable insights for designing efficient HAR systems. We finally discussed the findings and suggested principled design guidelines for future efficient HAR. To catalyze edge-HAR research, we open-source all materials in this work for future benchmarkingfootnote{https://github.com/zhaxidele/TinierHAR}

Problem

• Develop ultra-lightweight models for efficient human activity recognition on edge devices
• Balance accuracy and computational efficiency in resource-constrained wearable devices
• Provide systematic ablation study and design guidelines for efficient HAR systems

Innovation

• Ultra-lightweight deep learning architecture
• Residual depthwise separable convolutions and GRUs
• Systematic ablation study for spatial-temporal components