Deep neural network (DNN) shows great promise in providing more intelligence to ubiquitous end devices. However, the existing partition-offloading schemes adopt data-parallel or model-parallel collaboration between devices and the cloud, which does not make full use of the resources of end devices for deep-level parallel execution. This article proposes eDDNN (i.e., enabling Distributed DNN), a collaborative inference scheme over heterogeneous end devices using cross-platform Web technology, moving the computation close to ubiquitous end devices, improving resource utilization, and reducing the computing pressure of data centers. eDDNN implements D2D communication and collaborative inference among heterogeneous end devices with WebRTC protocol, divides the data and corresponding DNN model into pieces simultaneously, and then executes inference almost independently by establishing a layer dependency table. Besides, eDDNN provides a dynamic allocation algorithm based on deep reinforcement learning to minimize latency. We conduct experiments on various data sets and DNNs and further employ eDDNN into a mobile Web AR application to illustrate the effectiveness. The results show that eDDNN can achieve the latency decrease by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.98\times $ </tex-math></inline-formula> , reduce mobile energy by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1.8\times $ </tex-math></inline-formula> , and relieve the computing pressure of the edge server by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.57\times $ </tex-math></inline-formula> , against a typical partition-offloading approach. 

Enabling DNN Acceleration With Data and Model Parallelization Over Ubiquitous End Devices

The physical world we live in is accelerating digitalization with the vigorous development of Internet of Things (IoT). Following this trend, Web of Things (WoT) further enables fast and efficient creation of various applications that perceive and act on the physical world using standard Web technologies. A popular way for creating WoT applications is Trigger-Action Programming (TAP), which allows users to orchestrate the capabilities of IoT devices in the form of “if trigger, then action”. However, existing TAP approaches don’t support scenario-centric WoT applications which involve abstract modeling of physical environments and complex spatio-temporal dependencies between events and actions. In this paper, we propose an approach called SCTAP which supports Scenario-Centric Trigger-Action Programming based on software-defined physical environments. SCTAP defines a structured and conceptual representation for physical environments, which provides the required programming abstractions for WoT applications. Based on the representation, SCTAP defines a grammar for specifying scenario-centric WoT applications with spatio-temporal dependencies. Furthermore, we design a service-based architecture for SCTAP which supports the integration of device access, event perception, environment representation, and rule execution in a loosely-coupled and extensible way. We implement SCTAP as a WoT infrastructure and evaluate it with two case studies including a smart laboratory and a smart coffee house. The results confirm the usability, feasibility and efficiency of SCTAP and its implementation.

SCTAP: Supporting Scenario-Centric Trigger-Action Programming based on Software-Defined Physical Environments

Smart spaces such as smart homes deliver digital services to optimize space use and enhance user experience. They are composed of an Internet of Things (IoT), people, and physical content. They dif...

JinZhi

Papers

IoTranx: Transactions for Safer Smart Spaces