You Don't Need Attention: Gated Convolutional Modeling for Watch-Based Fall Detection
A new study presents Gated-CNN, a lightweight dual-stream architecture for fall detection using wearable devices. This model processes accelerometer and gyroscope data more efficiently than traditional self-attention mechanisms. The results show high accuracy in both offline and real-time evaluations, demonstrating its potential for practical applications in smartwatch-based fall detection.
- ▪The Gated-CNN model achieves average F1-scores of 93% to 97% across various datasets and real-time evaluations.
- ▪It utilizes a sigmoid gating module to enhance fall-discriminative features while suppressing background noise.
- ▪The model outperforms existing Transformer-based approaches in terms of computational efficiency and accuracy.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.20275 (cs) [Submitted on 19 May 2026] Title:You Don't Need Attention: Gated Convolutional Modeling for Watch-Based Fall Detection Authors:Sana Alamgeer, Ronish Kumar, Awatif Yasmin, Muhammad Irshad, Anne H. H. Ngu View a PDF of the paper titled You Don't Need Attention: Gated Convolutional Modeling for Watch-Based Fall Detection, by Sana Alamgeer and 4 other authors View PDF HTML (experimental) Abstract:Existing deep learning approaches for wearable fall detection systems rely on self-attention mechanisms that impose quadratic computational overhead, distributing weights across all time steps.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
