A Variational Autoencoder (VAE) is a generative deep learning model that encodes input data into a structured probabilistic latent space by learning the parameters of a distribution rather than a fixed point representation, then decodes samples from that distribution to reconstruct the original input. In WiFi CSI sensing and related domains, VAEs matter because they enable unsupervised or semi-supervised representation learning, allowing models to capture meaningful latent structure from unlabeled signal data and generate realistic synthetic samples to address data scarcity. Key variants relevant to these fields include conditional VAEs (CVAEs), which incorporate class or context labels to guide generation, and hybrid models that couple VAE-based generative components with physics-informed or contrastive learning objectives to improve the interpretability and transferability of learned representations.
Source Papers
- Context-Aware Predictive Coding: A Representation Learning Framework for WiFi Sensing ↗ — Context-Aware Predictive Coding: A Representation Learning F
- Deep Learning-Enhanced Human Sensing with Channel State Information: A Survey ↗ — Deep Learning-Enhanced Human Sensing with Channel State Info
- NeRF2: Neural Radio-Frequency Radiance Fields ↗ — NeRF2: Neural Radio-Frequency Radiance Fields
- Physics-Informed Deep Learning for Traffic State Estimation: A Survey and the Outlook ↗ — Physics-Informed Deep Learning for Traffic State Estimation:
- Recent trends in crowd analysis: A review ↗ — Recent trends in crowd analysis: A review
- Time matters: Empirical insights into the limits and challenges of temporal generalization in CSI-based Wi-Fi sensing ↗ — Time matters: Empirical insights into the limits and challen