MU-SHOT-Fi: Source-Free Domain Adaptation for Multi-User Wi-Fi Sensing
A group of researchers has introduced MU-SHOT-Fi, a framework for source-free unsupervised domain adaptation aimed at both single- and multi-user Wi-Fi sensing utilizing Channel State Information (CSI). This approach incorporates permutation-invariant set prediction alongside Hungarian matching throughout the source training phase, while also implementing a frozen-classifier backbone for adaptation in the target domain. To facilitate consistent adaptation in the absence of labels, the concept of occupancy-w has been introduced.
Key facts
- MU-SHOT-Fi is a source-free unsupervised domain adaptation framework.
- It addresses single- and multi-user Wi-Fi sensing.
- Uses permutation-invariant set prediction with Hungarian matching.
- Frozen-classifier backbone adaptation in target domain.
- Introduces occupancy-w for stable label-free adaptation.
- Published on arXiv with ID 2605.01369.
- Focuses on WiFi CSI-based human activity recognition (HAR).
- Aims to overcome domain shifts and CSI entanglement in multi-user settings.
Entities
Institutions
- arXiv