Real-Time AI Detection of Fast Radio Bursts: SETI's GPU-Streaming Pipeline with Holoscan and Edge Inference

To the point

Led by Andrew Siemion, the SETI team demonstrated real‑time AI detection of faint space signals by streaming telescope data into GPUs, with Luigi Cruz building the Holoscan‑based inference pipeline, Peter Ma training the AI model, and Wael Farah shaping the scientific framework in collaboration with Adam Thompson and Breakthrough Listen at Oxford, collecting over 90 billion data packets across 5 GHz in 15 hours and approaching 100 Gbps in real time, and planning to scale across a dozen sites and share the capability with international observatories and thousands of users, while noting the approach could be adapted to other signals and AI applications.