Abstract:Cross-domain generalization is an open problem in WiFi-based sensing due to variations in environments, devices, and subjects, causing domain shifts in channel state information. To address this, we propose Domain-Adversarial Test-Time Adaptation (DATTA), a novel framework combining domain-adversarial training (DAT), test-time adaptation (TTA), and weight resetting to facilitate adaptation to unseen target domains and to prevent catastrophic forgetting. DATTA is integrated into a lightweight, flexible architecture optimized for speed. We conduct a comprehensive evaluation of DATTA, including an ablation study on all key components using publicly available data, and verify its suitability for real-time applications such as human activity recognition. When combining a SotA video-based variant of TTA with WiFi-based DAT and comparing it to DATTA, our method achieves an 8.1% higher F1-Score. The PyTorch implementation of DATTA is publicly available at: <a class="link-external link-https" href="https://github.com/StrohmayerJ/DATTA" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the generalization problem in WiFi - based cross - domain human activity recognition (HAR). Specifically, due to differences in environments, devices, and subjects, WiFi channel state information (CSI) varies significantly between different domains, causing trained models to struggle to maintain high performance in unseen environments. This cross - domain generalization problem is one of the main challenges faced by WiFi - sensing technologies in practical applications. To address this issue, the authors propose a new framework named **Domain - Adversarial Test - Time Adaptation (DATTA)**. DATTA combines the following methods: 1. **Domain - Adversarial Training (DAT)**: Through adversarial learning, the model can learn domain - independent features, thereby improving cross - domain generalization ability. 2. **Test - Time Adaptation (TTA)**: Fine - tune the model during the inference stage so that it can adapt to the data distribution of new domains without additional labeled data. 3. **Weight Resetting**: By randomly resetting some model parameters, prevent the model from forgetting the learned domain - invariant features during long - term adaptation and avoid catastrophic forgetting. ### Specific contributions of the paper 1. **Propose the DATTA framework**: Combine domain - adversarial training with test - time adaptation to achieve real - time, robust cross - domain generalization. 2. **Design a lightweight architecture**: Optimize computational efficiency to ensure that the model is suitable for real - time applications. 3. **Comprehensive evaluation and ablation study**: Conducted exhaustive experimental validations using public datasets, analyzed the functions of each component, and proved its effectiveness in practical applications. ### Summary This paper presents an innovative solution - the DATTA framework - in response to the challenges of WiFi - sensing technologies in cross - domain generalization. By combining domain - adversarial training and test - time adaptation and introducing a weight - resetting mechanism, DATTA can maintain high recognition accuracy in different environments, overcome the limitations of existing methods, and provide strong support for the practical applications of WiFi - based HAR.

DATTA: Domain-Adversarial Test-Time Adaptation for Cross-Domain WiFi-Based Human Activity Recognition

Privacy-Preserving and Cross-Domain Human Sensing by Federated Domain Adaptation with Semantic Knowledge Correction

A Semi-Supervised Transfer Learning with Dynamic Associate Domain Adaptation for Human Activity Recognition Using WiFi Signals

Wi-Adaptor: Fine-grained Domain Adaptation in WiFi-based Activity Recognition

Adversarial AI applied to cross-user inter-domain and intra-domain adaptation in human activity recognition using wireless signals

Weakly Supervised Test-Time Domain Adaptation for Object Detection

Dual-Adversarial Domain Adaptation for Generalized Replay Attack Detection

Target-oriented Semi-supervised Domain Adaptation for WiFi-based HAR

Data Augmentation Techniques for Cross-Domain WiFi CSI-based Human Activity Recognition

Deep Adaptive Temporal Pooling for Activity Recognition

Unsupervised Domain Adaptation for Activity Recognition Across Heterogeneous Datasets

Optimization-Free Test-Time Adaptation for Cross-Person Activity Recognition

ADL-ID: Adversarial Disentanglement Learning for Wireless Device Fingerprinting Temporal Domain Adaptation

Test-time Adaptation in the Dynamic World with Compound Domain Knowledge Management

A Pairwise Attentive Adversarial Spatiotemporal Network for Cross-Domain Few-Shot Action Recognition-R2.

Object-based (yet Class-agnostic) Video Domain Adaptation

Test-Time Training-Free Domain Adaptation

Bi-TTA: Bidirectional Test-Time Adapter for Remote Physiological Measurement

Domain Adaptation Under Behavioral and Temporal Shifts for Natural Time Series Mobile Activity Recognition

Temporal Attentive Alignment for Large-Scale Video Domain Adaptation

MLFA: Toward Realistic Test Time Adaptive Object Detection by Multi-Level Feature Alignment