Research map / Time series / Foundation models for time series
DeepSenseMoE: Harnessing Power of Time Series Foundation Models for Few-Shot Human Activity Recognition
Zenan Fu, Dongzhou Cheng, Lei Zhang, Wenbo Huang, Zhenghao Chen, Hao Wu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add multi-scale convolutional experts with router (case 2 vs case 1) (ablation) | DeepSenseMoE · MotionSense 1-shot | accuracy (%): 76.9 -> 80.1 | Improved |
| Add shared expert isolation (ablation) | DeepSenseMoE · MotionSense 1-shot | accuracy (%): 80.1 -> 81.3 | Improved |
| Add hierarchical supervised contrastive alignment (ablation) | DeepSenseMoE · MotionSense 1-shot | accuracy (%): 81.3 -> 83.1 | Improved |
| Use convolutional experts instead of linear experts (ablation) | DeepSenseMoE · MotionSense few-shot | accuracy (%) | Improved |
| Use supervised contrastive instead of Dispersive or InfoNCE (ablation) | DeepSenseMoE · MotionSense 5-shot | accuracy (%): 90.2 -> 91.7 | Improved |
| Use DeepSenseMoE instead of Adapter | MOMENT-small · MotionSense 1-shot | accuracy (%): 68.6 -> 83.1 | Improved |
| Use DeepSenseMoE instead of LoRA | MOMENT-small · MotionSense 1-shot | accuracy (%): 65.6 -> 83.1 | Improved |
| Use DeepSenseMoE instead of full fine-tuning | MOMENT-small · PAMAP2 1-shot | accuracy (%): 59.6 -> 74.4 | Improved |
| Use DeepSenseMoE instead of Vi2ACT on HHAR 1-shot | DeepSenseMoE · HHAR 1-shot | accuracy (%): 75.5 -> 75.1 | Hurt |
| Use DeepSenseMoE instead of Vi2ACT on MotionSense 5-shot | DeepSenseMoE · MotionSense 5-shot | accuracy (%): 82.2 -> 91.7 | Improved |
| Use DeepSenseMoE compared to FT-Head on MotionSense full supervision | DeepSenseMoE · MotionSense Full | accuracy (%): 98.0 -> 98.9 | Improved |
| Vary lambda1 and lambda2 (ablation) | DeepSenseMoE · MotionSense 5-shot | accuracy (%) | No clear effect |
Closest papers by meaning
- UniMTS: Unified Pre-training for Motion Time Series (NeurIPS 2024 · similarity 0.87)
- Scalable Hierarchical Deep Neural Network for Time Series Analysis in Wearable Sensor-based Human Activity Recognition (AAAI 2025 · similarity 0.87)
- Scaling Wearable Foundation Models (ICLR 2025 · similarity 0.87)
- Decomposing and Fusing Intra- and Inter-Sensor Spatio-Temporal Signal for Multi-Sensor Wearable Human Activity Recognition (AAAI 2025 · similarity 0.85)
- MoPFormer: Motion-Primitive Transformer for Wearable-Sensor Activity Recognition (NeurIPS 2025 · similarity 0.85)
- HARDVS: Revisiting Human Activity Recognition with Dynamic Vision Sensors (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
