Over-the-Air Ensemble Inference with Model Privacy

Name: Over-the-Air Ensemble Inference with Model Privacy
Uploaded: 2026-10-23T08:57:18.553923+00:00
Duration: 438 s
Description: This work introduces an over-the-air ensemble inference framework for wireless edge devices that enables bandwidth-efficient and private distributed classification. It uses multiple independently trained models and exploits channel gains to reduce power consumption through random client participation, ensuring privacy by embedding Gaussian noise directly into the predictions.

This work introduces an over-the-air ensemble inference framework for wireless edge devices that enables bandwidth-efficient and private distributed classification. It uses multiple independently trained models and exploits channel gains to reduce power consumption through random client participation, ensuring privacy by embedding Gaussian noise directly into the predictions.

Open the full episode