Research map / Detection and segmentation / Open-set and few-shot detection
Dywave: Event-Aligned Dynamic Tokenization for Heterogeneous IoT Sensing Signals
Tomoyoshi Kimura, Denizhan Kara, Jinyang Li, Hongjue Zhao, Yigong Hu, Yizhuo Chen, Xiaomin Ouyang, Shengzhong Liu, Tarek F. Abdelzaher
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 |
|---|---|---|---|
| replace uniform patching with Dywave event-aligned tokenization | Mamba2 backbone · PAMAP2 / RWHAR accelerometer | accuracy: RWHAR 0.7340 -> 0.8517 | Improved |
| add hierarchical wavelet embedding (vs PatchTST embedding) (ablation) | Dywave · MOD audio | accuracy: 0.8520 -> 0.9002 | Improved |
| add reconstruction objective (ablation) | Dywave · MOD audio | accuracy: 0.7823 -> 0.9002 | Improved |
| use similarity-based saliency instead of CNN boundary prediction (ablation) | Dywave · MOD audio | accuracy: 0.8064 -> 0.9002 | Improved |
| add temporal fusion of non-anchor segments (ablation) | Dywave · MOD audio | accuracy: 0.7582 -> 0.9002 | Improved |
| add temporal fusion of non-anchor segments (ablation) | Dywave · MOD seismic | accuracy: 0.7863 vs 0.7930 | No clear effect |
| use embedding-space cosine saliency instead of spectral energy (ablation) | Dywave · MOD audio | accuracy: 0.8533 -> 0.9002 | Improved |
| use Dywave tokens instead of PatchTST under head-only finetuning | Transformer · MOD-Seismic cross-domain | accuracy: 0.3414 -> 0.7565 | Improved |
| use Dywave tokens instead of PatchTST under full finetuning | Transformer · MOD-Seismic cross-domain | accuracy: 0.8111 -> 0.8119 | No clear effect |
| use Dywave instead of dynamic patching baselines LightGTS and Ruptures | Transformer · RWHAR | accuracy: 0.8586 -> 0.9094 | Improved |
| use Dywave tokens instead of WaveToken discretization baseline | Transformer · MOD seismic | accuracy: 0.5358 -> 0.7917 | Improved |
| use Dywave instead of fixed patching, token count | Transformer · Ego4D 5s to 30s | number of tokens: PatchTST 59 -> 372 vs Dywave 15.5 -> 35 | Improved |
Closest papers by meaning
- Fine-grained Control of Generative Data Augmentation in IoT Sensing (NeurIPS 2024 · similarity 0.83)
- CSI-Bench: A Large-Scale In-the-Wild Dataset for Multi-task WiFi Sensing (NeurIPS 2025 · similarity 0.81)
- SensorLM: Learning the Language of Wearable Sensors (NeurIPS 2025 · similarity 0.80)
- WiFi CSI Based Temporal Activity Detection via Dual Pyramid Network (AAAI 2025 · similarity 0.79)
- MoPFormer: Motion-Primitive Transformer for Wearable-Sensor Activity Recognition (NeurIPS 2025 · similarity 0.79)
- Decomposing and Fusing Intra- and Inter-Sensor Spatio-Temporal Signal for Multi-Sensor Wearable Human Activity Recognition (AAAI 2025 · similarity 0.79)
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.
