Transferable Deployment of Semantic Edge Inference Systems via Unsupervised Domain Adaption

TL;DR

This work tackles the challenge of deploying semantic edge inference systems across environments with varying data and wireless channel conditions. It proposes DASEIN, a two-step transfer method combining unsupervised domain adaptation to align cross-domain data distributions and knowledge distillation to adapt to different target-channel conditions without labeled target data. The approach demonstrates significant gains over baselines in cross-domain object recognition tasks and shows that enabling digital transmissions can further improve performance under noise. The results underscore the practical value of swift, label-free transfer deployment for edge inference in dynamic IoT scenarios, with insights on bandwidth-accuracy tradeoffs and channel-aware adaptation strategies.

Abstract

This paper investigates deploying semantic edge inference systems for performing a common image clarification task. In particular, each system consists of multiple Internet of Things (IoT) devices that first locally encode the sensing data into semantic features and then transmit them to an edge server for subsequent data fusion and task inference. The inference accuracy is determined by efficient training of the feature encoder/decoder using labeled data samples. Due to the difference in sensing data and communication channel distributions, deploying the system in a new environment may induce high costs in annotating data labels and re-training the encoder/decoder models. To achieve cost-effective transferable system deployment, we propose an efficient Domain Adaptation method for Semantic Edge INference systems (DASEIN) that can maintain high inference accuracy in a new environment without the need for labeled samples. Specifically, DASEIN exploits the task-relevant data correlation between different deployment scenarios by leveraging the techniques of unsupervised domain adaptation and knowledge distillation. It devises an efficient two-step adaptation procedure that sequentially aligns the data distributions and adapts to the channel variations. Numerical results show that, under a substantial change in sensing data distributions, the proposed DASEIN outperforms the best-performing benchmark method by 7.09% and 21.33% in inference accuracy when the new environment has similar or 25 dB lower channel signal to noise power ratios (SNRs), respectively. This verifies the effectiveness of the proposed method in adapting both data and channel distributions in practical transfer deployment applications.

Figures (14)

• Figure 1: Two distinct deployment scenarios of semantic edge inference system for object detection.
• Figure 2: The schematics of the considered edge inference system.
• Figure 3: The simplified flowchart of DASEIN system. Step 1 mitigates the variations in data distribution by training model $G$ through UDA. Step 2 further mitigates the simultaneous variations of channel distributions by utilizing G as the teacher model $G_{tc}$ to train a new student model $G_{st}$.
• Figure 4: Step 1 mitigates the variations in data distribution by minimizing cross-entropy loss $\mathcal{L}_{CE}$ and UDA loss $\mathcal{L}_{UDA}$, whereas step 2 further mitigates the simultaneous variations of channel distributions by minimizing $\mathcal{L}_{CE}$, $\mathcal{L}_{UDA}$, and knowledge distillation loss $\mathcal{L}_{KD}$.
• Figure 5: The digital schematics of the considered edge inference system.