AirNet: Neural Network Transmission over the Air

TL;DR

AirNet introduces a novel approach for efficiently transmitting deep neural networks over wireless channels by directly mapping network parameters to channel symbols, optimizing for accuracy under power and latency constraints.

Contribution

The paper presents a new joint source-channel coding method for DNN transmission, including training strategies for robustness, pruning, expansion, and multi-condition training.

Findings

AirNet outperforms separation-based methods in accuracy.

Pruning and expansion improve robustness and bandwidth utilization.

Training multiple DNNs for different conditions reduces memory requirements.

Abstract

State-of-the-art performance for many edge applications is achieved by deep neural networks (DNNs). Often, these DNNs are location- and time-sensitive, and must be delivered over a wireless channel rapidly and efficiently. In this paper, we introduce AirNet, a family of novel training and transmission methods that allow DNNs to be efficiently delivered over wireless channels under stringent transmit power and latency constraints. This corresponds to a new class of joint source-channel coding problems, aimed at delivering DNNs with the goal of maximizing their accuracy at the receiver, rather than recovering them with high fidelity. In AirNet, we propose the direct mapping of the DNN parameters to transmitted channel symbols, while the network is trained to meet the channel constraints, and exhibit robustness against channel noise. AirNet achieves higher accuracy compared to separation-based alternatives. We further improve the performance of AirNet by pruning the network below the available bandwidth, and expanding it for improved robustness. We also benefit from unequal error protection by selectively expanding important layers of the network. Finally, we develop an approach, which simultaneously trains a spectrum of DNNs, each targeting a different channel condition, resolving the impractical memory requirements of training distinct networks for different channel conditions.