Research map / Label-efficient and robust learning / Test-time adaptation
WiTTA-Bench: Benchmarking Test-Time Adaptation for WiFi Sensing
Bing Li, Qiang Wang, Junda Lu, Le Zhang, Yun Liu, Ce Zhu, Wei Cui
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 |
|---|---|---|---|
| use lightweight normalization/entropy OTTA (DUA, EATA, TENT) instead of consistency-based CoTTA | OTTA on WiFi CSI · WiHAR-Dual | avg cross-domain accuracy (%): 26.2 -> 35.7 (CoTTA vs TENT) | Improved |
| use consistency-based OTTA instead of normalization-based OTTA | OTTA on WiFi CSI · WiHAR-Dual and CSLOS | accuracy | Hurt |
| use TTDA instead of OTTA for cross-device shift | ASFA vs OTTA · Cross-device | accuracy: up to +30% absolute (ASFA) | Improved |
| use OTTA for cross-device shift | OTTA methods · Cross-device | accuracy | No clear effect |
| apply TENT on ResNet-10 | TENT · WiHAR-Dual cross-device | accuracy (%): 14.29 -> 28.55 | Improved |
| apply T3A on ResNet-10 cross-device | T3A · WiHAR-Dual cross-device | accuracy (%): 14.29 -> 14.29 | No clear effect |
| tune learning rate and batch size away from optimum (ablation) | ASFA (TTDA) · WiHAR-Dual | accuracy (%): ~63% -> 35% | Hurt |
| vary learning rate and batch size (ablation) | TENT/DUA/EATA (OTTA) · CSLOS + WiHAR-Dual | accuracy | No clear effect |
| use stronger source model (ablation) | TTA methods · CSLOS and WiHAR-Dual E1 to E2 | TTA accuracy | Improved |
| use higher source accuracy for SHOT (ablation) | SHOT · WiHAR-Dual | TTA accuracy | Hurt |
| predict TTA gain from feature-shift metrics (SS, CH, DB, MMD) (ablation) | domain shift metrics · WiHAR-Dual | Spearman correlation with accuracy: r from -0.21 to 0.18 | No clear effect |
| use clustering/self-supervised TTDA (ASFA, SHOT++) over pseudo-label ASL, DaC | TTDA · WiHAR-Dual and CSLOS | avg cross-domain accuracy | Improved |
Closest papers by meaning
- CSI-Bench: A Large-Scale In-the-Wild Dataset for Multi-task WiFi Sensing (NeurIPS 2025 · similarity 0.87)
- WiMANS: A Benchmark Dataset for WiFi-based Multi-user Activity Sensing (ECCV 2024 · similarity 0.85)
- ConSense: Continually Sensing Human Activity with WiFi via Growing and Picking (AAAI 2025 · similarity 0.83)
- WiFi CSI Based Temporal Activity Detection via Dual Pyramid Network (AAAI 2025 · similarity 0.82)
- TinyTTA: Efficient Test-time Adaptation via Early-exit Ensembles on Edge Devices (NeurIPS 2024 · similarity 0.81)
- SNAP: Low-Latency Test-Time Adaptation with Sparse Updates (NeurIPS 2025 · similarity 0.81)
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.
