Distributed Deep Convolutional Neural Networks for the Internet-of-Things
📝 Original Paper Info
- Title: Distributed Deep Convolutional Neural Networks for the Internet-of-Things- ArXiv ID: 1908.01656
- Date: 2021-07-30
- Authors: Simone Disabato, Manuel Roveri, Cesare Alippi
📝 Abstract
Severe constraints on memory and computation characterizing the Internet-of-Things (IoT) units may prevent the execution of Deep Learning (DL)-based solutions, which typically demand large memory and high processing load. In order to support a real-time execution of the considered DL model at the IoT unit level, DL solutions must be designed having in mind constraints on memory and processing capability exposed by the chosen IoT technology. In this paper, we introduce a design methodology aiming at allocating the execution of Convolutional Neural Networks (CNNs) on a distributed IoT application. Such a methodology is formalized as an optimization problem where the latency between the data-gathering phase and the subsequent decision-making one is minimized, within the given constraints on memory and processing load at the units level. The methodology supports multiple sources of data as well as multiple CNNs in execution on the same IoT system allowing the design of CNN-based applications demanding autonomy, low decision-latency, and high Quality-of-Service.💡 Summary & Analysis
This paper addresses the challenges of deploying deep learning (DL) solutions in Internet-of-Things (IoT) units that are constrained by limited memory and computational power. These constraints often prevent the execution of typical DL models, which require substantial memory and processing capabilities. The authors propose a design methodology for distributing Deep Convolutional Neural Networks (CNNs) across IoT systems to support real-time data processing and decision-making while respecting these constraints.The core technology introduced in this paper formalizes the deployment strategy as an optimization problem aimed at minimizing latency between data collection and decision-making phases, within given memory and processing load limits. The methodology supports multiple data sources and multiple CNN models operating simultaneously on a single IoT system, which is crucial for applications requiring autonomy, low-latency decisions, and high-quality service.
The results demonstrate the effectiveness of this approach in various experimental settings, particularly with a 5-layer CNN example. By successfully managing memory and processing loads efficiently, the methodology enables real-time execution and decision-making processes that are essential for IoT applications.
This research is significant as it paves the way for more efficient deployment of DL models on constrained IoT devices, potentially expanding their applicability across various industries where real-time data analysis and quick decisions are critical.