SenseFi: A Library and Benchmark on Deep-Learning-Empowered WiFi Human Sensing (2207.07859v3)

Abstract: WiFi sensing has been evolving rapidly in recent years. Empowered by propagation models and deep learning methods, many challenging applications are realized such as WiFi-based human activity recognition and gesture recognition. However, in contrast to deep learning for visual recognition and natural language processing, no sufficiently comprehensive public benchmark exists. In this paper, we review the recent progress on deep learning enabled WiFi sensing, and then propose a benchmark, SenseFi, to study the effectiveness of various deep learning models for WiFi sensing. These advanced models are compared in terms of distinct sensing tasks, WiFi platforms, recognition accuracy, model size, computational complexity, feature transferability, and adaptability of unsupervised learning. It is also regarded as a tutorial for deep learning based WiFi sensing, starting from CSI hardware platform to sensing algorithms. The extensive experiments provide us with experiences in deep model design, learning strategy skills and training techniques for real-world applications. To the best of our knowledge, this is the first benchmark with an open-source library for deep learning in WiFi sensing research. The benchmark codes are available at https://github.com/xyanchen/WiFi-CSI-Sensing-Benchmark.

• The paper introduces SenseFi, the first benchmark library that rigorously evaluates deep learning models on WiFi CSI-based human sensing tasks.
• It compares several architectures, including CNNs, GRUs, and Transformers, and finds that CNNs effectively balance recognition accuracy and computational efficiency.
• It demonstrates the potential of transfer and unsupervised learning strategies, paving the way for data-efficient models and more robust real-world applications.

Overview of SenseFi: A Benchmark on Deep-Learning-Enhanced WiFi Human Sensing

The paper "SenseFi: A Library and Benchmark on Deep-Learning-Empowered WiFi Human Sensing" provides an extensive analysis of deep learning techniques applied to WiFi-based human sensing. The authors propose SenseFi, a comprehensive benchmark designed to evaluate and compare various deep learning models in the context of WiFi sensing tasks. This work addresses the lack of a unified benchmark for this domain and explores the efficacy of multiple deep learning architectures, including CNNs, RNNs, LSTMs, and Transformers, across different WiFi platforms.

Key Contributions and Findings

1. Benchmark Development: The authors introduce SenseFi, the first open-source library and benchmark for WiFi CSI (Channel State Information) sensing using deep learning. This initiative provides a structured foundation for evaluating machine learning models in WiFi sensing applications.
2. Deep Learning Architectures: The paper evaluates various models, such as MLP, CNN, GRU, LSTM, and Transformers, on multiple datasets, analyzing their performance in terms of recognition accuracy, model size, and computational complexity. The CNN and GRU models exhibit a strong balance of accuracy and efficiency, particularly favoring CNNs for their transferability in diverse tasks.
3. Datasets and Platforms: Four CSI datasets are utilized, including public datasets like UT-HAR and Widar, as well as newly collected datasets NTU-Fi HAR and NTU-Fi Human-ID, which leverage different WiFi platforms. This diversity underlines the adaptability of the benchmark across various scenarios and measurement tools.
4. Learning Strategies: The paper evaluates not only supervised learning but also transfer learning and unsupervised learning. Transfer learning effectively leverages knowledge across similar tasks, whereas unsupervised learning demonstrates substantial potential in initializing models for enhanced generalization.
5. Challenges and Future Directions: The paper highlights ongoing challenges such as data-efficient learning, model compression, multi-modal learning, and the importance of cross-modal approaches. These areas are crucial for progressing toward more practical and robust applications in real-world settings.

Implications and Future Prospects

The development of SenseFi sets a foundational benchmark that facilitates the rigorous evaluation of machine learning models in WiFi-based human sensing. The comprehensive analysis provided in this paper can guide future research, especially in improving model efficiency and deployability in edge devices. The emphasis on unsupervised learning and transfer learning opens pathways for advancements in data-efficient methods, crucial for reducing the data annotation burden in large-scale applications.

Future developments in this area are expected to focus on integrating more adaptive learning strategies, enhancing model interpretability, and addressing security challenges to ensure trustworthy and practical deployment in smart environments.

In conclusion, this paper effectively fills a significant gap in WiFi sensing research by furnishing a robust benchmark and offering valuable insights for further exploration in AI-driven human sensing applications.