Wi-Fi CSI-Based Human Activity Recognition with Causal Interpretability
A new study has unveiled a method for recognizing human activities using Wi-Fi Channel State Information (CSI) combined with deep learning and causal interpretability. This innovative approach utilizes a categorical variational autoencoder, which employs Gumbel-Softmax latent variables to transform raw CSI data into discrete forms. The encoder is then fixed to create a clear mapping to one-hot latent trajectories. By analyzing these trajectories, researchers can build class-conditional temporal dependency graphs. This technique aims to enhance predictive accuracy and provide symbolic control, addressing the limitations of unclear continuous latent representations and traditional symbolic methods that struggle with raw CSI data. You can check out the research on arXiv under the identifier 2604.22979.
Key facts
- The paper addresses Human Activity Recognition (HAR) using Wi-Fi Channel State Information (CSI).
- It proposes a fully automatic and strictly decoupled pipeline.
- CSI magnitude windows are compressed by a categorical variational autoencoder with Gumbel-Softmax latent variables.
- The encoder is frozen and used as a deterministic mapping to one-hot latent trajectories.
- Causal discovery is performed on these trajectories to estimate class-conditional temporal dependency graphs.
- The approach aims for causal interpretability, symbolic controllability, and direct operation on high-dimensional raw signals.
- Deep neural models achieve strong predictive performance but rely on opaque continuous latent representations.
- Purely symbolic approaches cannot process raw CSI streams.
- The paper is published on arXiv with identifier 2604.22979.
Entities
Institutions
- arXiv