Neuromorphic Wireless Device-Edge Co-Inference via the Directed Information Bottleneck

TL;DR

This work tackles device-edge co-inference by placing the sensing and feature extraction on a neuromorphic device and delegating the heavy edge computation to a conventional server. It introduces a transmitter-centric encoder design based on the directed information bottleneck, optimized via a variational bound (S-VDIB) and Monte Carlo gradient methods, while keeping the edge decoder separate. The approach is validated on neuromorphic vision datasets (MNIST-DVS and N-MNIST), showing superior end-to-end task performance under limited communication budgets and robustness to SNR variability, compared to traditional SSCC and JSCC baselines. A preliminary testbed with a neuromorphic camera and impulse-radio link demonstrates practical feasibility and sets the stage for end-to-end learning over wireless neuromorphic systems in robotics and biomedical applications.

Abstract

An important use case of next-generation wireless systems is device-edge co-inference, where a semantic task is partitioned between a device and an edge server. The device carries out data collection and partial processing of the data, while the remote server completes the given task based on information received from the device. It is often required that processing and communication be run as efficiently as possible at the device, while more computing resources are available at the edge. To address such scenarios, we introduce a new system solution, termed neuromorphic wireless device-edge co-inference. According to it, the device runs sensing, processing, and communication units using neuromorphic hardware, while the server employs conventional radio and computing technologies. The proposed system is designed using a transmitter-centric information-theoretic criterion that targets a reduction of the communication overhead, while retaining the most relevant information for the end-to-end semantic task of interest. Numerical results on standard data sets validate the proposed architecture, and a preliminary testbed realization is reported.

Figures (7)

• Figure 1: Neuromorphic wireless device-edge co-inference.
• Figure 2: The proposed hybrid neuromorphic-classical wireless device-edge co-inference solution encompassing an encoding SNN at the transmitter and a decoding ANN at the receiver with IR-based transmission.
• Figure 3: Test error rate as a function of the SNR per bit, $E_b / N_0$, for the two standard datasets MNIST-DVS (left) and N-MNIST (right).
• Figure 4: Test error rate for S-VDIB on the MNIST-DVS dataset with an SNR mismatch between training and testing conditions ($k=256$).
• Figure 5: (top) Test classification error and (bottom) per-neuron spike rate as a function of the training epoch number for the MNIST-DVS dataset ($k=256$, $E_b/N_0=-4.7$ dB).