We study speculative ensembles for RF classificationwith a fast model that accepts confident inputs and a slow modelthat arbitrates the remainder; predictions are fused by confidence weighted probabilities. Under a 50 ms budget we attain 92.2%with median latency 26.0 ms—a 1.65× speed-up vs the slow model(43.0 ms) while retaining most of its accuracy (92.8%).Index Terms—RF … Continue reading Speculative Ensembles for Real-Time RF Classification