OrcoDCS: An IoT-Edge Orchestrated Online Deep Compressed Sensing Framework

TL;DR

Compressed data aggregation in wireless sensor networks is hindered by limited adaptability to different sensing tasks and environmental changes, particularly for follow-up DL applications. OrcoDCS proposes an IoT-Edge orchestrated online training framework using a specially designed asymmetric autoencoder to shift training work from IoT devices to edge servers and to inject Gaussian noise for robustness. The encoder is shallow and located at the data aggregator, while the decoder remains on the edge server, enabling online, task-specific adaptation with reduced device overhead and flexible latent dimensions. Experiments on MNIST and GTSRB show OrcoDCS achieves up to 10x transmission savings and superior reconstruction quality and downstream classifier performance compared with state-of-the-art offline DCDA baselines.

Abstract

Compressed data aggregation (CDA) over wireless sensor networks (WSNs) is task-specific and subject to environmental changes. However, the existing compressed data aggregation (CDA) frameworks (e.g., compressed sensing-based data aggregation, deep learning(DL)-based data aggregation) do not possess the flexibility and adaptivity required to handle distinct sensing tasks and environmental changes. Additionally, they do not consider the performance of follow-up IoT data-driven deep learning (DL)-based applications. To address these shortcomings, we propose OrcoDCS, an IoT-Edge orchestrated online deep compressed sensing framework that offers high flexibility and adaptability to distinct IoT device groups and their sensing tasks, as well as high performance for follow-up applications. The novelty of our work is the design and deployment of IoT-Edge orchestrated online training framework over WSNs by leveraging an specially-designed asymmetric autoencoder, which can largely reduce the encoding overhead and improve the reconstruction performance and robustness. We show analytically and empirically that OrcoDCS outperforms the state-of-the-art DCDA on training time, significantly improves flexibility and adaptability when distinct reconstruction tasks are given, and achieves higher performance for follow-up applications.

Figures (8)

• Figure 1: The OrcoDCS architecture. First, IoT devices send raw sensing data $X$ to the data aggregator through performing the intra-cluster raw data aggregation (➊). Next, the data aggregator and the edge server train an asymmetric autoencoder using the training procedure of the IoT-Edge orchestrated asymmetric autoencoder (➋). Once the training procedure finishes, IoT devices can send compressed data $Y$ to the data aggregator through the data aggregation of OrcoDCS over IoT networks (➌).
• Figure 2: Reconstruction results of OrcoDCS and DCSNet for three digits in MNIST (upper line) and three traffic signs in GTSRB (lower line). Clearly, the reconstruction results produced by OrcoDCS are much clearer and more similar to the original images when compared to those generated by DCSNet.
• Figure 3: Transmission cost for OrcoDCS and DCSNet. OrcoDCS can save up to $10\times$ transmission cost than DCSNet.
• Figure 4: Breakdown of time-to-loss performance. OrcoDCS can achieve lower loss faster than DCSNet in terms of training time.
• Figure 5: Breakdown of classifier performance using the data reconstructed by ours and DCSNet. OrcoDCS can achieve higher classification performance than DCSNet.