Contrastive learning is a self-supervised representation learning approach that trains models to produce similar embeddings for augmented or semantically related views of the same input while pushing apart representations of dissimilar samples, enabling useful feature extraction without requiring labeled data. In WiFi CSI sensing, this matters because labeled CSI data is costly and environment-specific, and contrastive learning allows models to learn transferable, generalizable representations from unlabeled signals across diverse environments, users, and hardware configurations. Key variants applied in this domain include Contrastive Predictive Coding (CPC), which extends the principle to the temporal dimension by predicting future latent representations from past context, and context-aware formulations such as CAPC that further incorporate environmental context to improve cross-domain robustness.
Source Papers
- A Survey on Wi-Fi Sensing Generalizability: Taxonomy, Techniques, Datasets, and Future Research Prospects ↗ — A Survey on Wi-Fi Sensing Generalizability: Taxonomy, Techni
- Context-Aware Predictive Coding: A Representation Learning Framework for WiFi Sensing ↗ — Context-Aware Predictive Coding: A Representation Learning F