Dywave: Dynamic Tokenization Framework for IoT Sensing Signals
A new framework called Dywave proposes dynamic tokenization for heterogeneous IoT sensing signals. The method uses wavelet-based hierarchical decomposition to align token representations with intrinsic temporal structures and underlying physical events. It identifies meaningful temporal boundaries corresponding to semantic events and adaptively compresses redundant intervals while preserving temporal coherence. Evaluations on five real-world IoT sensing datasets covering activity recognition, stress assessment, and nearby object detection demonstrate its effectiveness. The paper is available on arXiv under identifier 2605.14014.
Key facts
- Dywave is a dynamic tokenization framework for IoT sensing signals.
- It uses wavelet-based hierarchical decomposition.
- It identifies temporal boundaries corresponding to semantic events.
- It adaptively compresses redundant intervals while preserving temporal coherence.
- Evaluated on five real-world IoT sensing datasets.
- Datasets cover activity recognition, stress assessment, and nearby object detection.
- Paper available on arXiv: 2605.14014.
- IoT systems collect heterogeneous sensing signals for intelligent applications.
Entities
Institutions
- arXiv