LITEWAY: LIghtweight HAR via Temporal Efficient highWAY

2026-08-10

AI Summary

This work addresses the computational and energy constraints of deploying deep models for human activity recognition (HAR) on wearable devices, where existing lightweight approaches often rely on recurrent architectures that hinder parallelization and incur high inference latency. To overcome these limitations, we propose a modality-agnostic, fully convolutional framework for multichannel temporal modeling that eliminates recurrent structures entirely. Our method efficiently captures temporal dependencies through structured convolutional decomposition, lightweight convolutional blocks, strided temporal convolutions, and a novel convolution-attention pooling mechanism. This approach achieves state-of-the-art performance with macro F1 scores comparable to advanced methods across 16 benchmark datasets, while reducing model size by 3.87–9.52× and deployment energy consumption by 1.46–3.14×—marking the first successful realization of efficient fully convolutional modeling in lightweight HAR.

Abstract

Wearable human activity recognition (HAR) remains challenging due to the computational and energy constraints of deep learning models on resource-limited devices. Existing lightweight approaches often rely on recurrent architectures (e.g., GRU and LSTM), limiting parallelism and increasing inference latency. We propose LITEWAY, a modality-agnostic, fully convolutional framework for multichannel sensor time series that replaces recurrent temporal modeling with structured convolutional decomposition. LITEWAY combines lightweight convolutional blocks, strided temporal processing, and convolution-attention pooling to efficiently capture temporal dependencies while reducing computational complexity. We evaluate LITEWAY on 16 HAR datasets against TinyHAR, TinierHAR, and MLP-HAR. LITEWAY achieves competitive macro F1 while reducing model size by 4.06x-9.52x (Light) and 3.87x-9.07x (Full) compared with TinyHAR and TinierHAR. Deployment experiments further show energy reductions of 2.29x-3.14x (Light) and 1.46x-2.01x (Full) compared with TinierHAR and MLP-HAR, highlighting efficient fully convolutional temporal modeling for wearable HAR. The source code is publicly available at https://github.com/dominique-nshimyimana/liteway.

Problem

• Human Activity Recognition
• Wearable Devices
• Computational Constraints
• Energy Efficiency
• Lightweight Models

Innovation

• lightweight HAR
• fully convolutional
• temporal modeling
• convolutional decomposition
• energy efficiency