WiFo-2 scalable 6G/ 7G AI foundation model generalizes across 9 communication/ sensing tasks using 1% of labeled training data

https://www.eurekalert.org/news-releases/1143014

"unifies fragmented communications/ sensing physical-layer design for heterogeneous networks... built on space-time-frequency dataset containing 11.6B channel state information points, combines masked denoising autoencoder with dynamic sparse mixture-of-experts mechanism... 0-shot channel reconstruction of task-specific models on unseen scenarios, tested over-the-air on software-defined radio hardware... single pretrained model capable of replacing fragmented wireless AI pipelines"