Description
A Generative Adversarial Network trains a generator and a discriminator in a min-max game so the generator learns to produce samples indistinguishable from real data. In CSI sensing GANs appear in three roles: (1) data-augmentation for rare classes (falls, multi-person scenes), (2) cross-domain CSI translation for domain-adaptation, and (3) anomaly detection via reconstruction error. They are conceptually distinct from adversarial-training despite sharing the word.
When it's used
- Synthetic CSI generation for rare-class augmentation
- Cross-environment CSI translation
- CSI anomaly detection via discriminator scores
Limitations
- Training instability — mode collapse is common
- Synthetic CSI quality is hard to evaluate
- Outperformed by diffusion models on some generation benchmarks