[Paper Review] Distributing Intelligence to the Edge and Beyond

Edgar Ramos, Roberto Morabito+1 more|arXiv (Cornell University)|Jul 25, 2019

Advanced Malware Detection Techniques22 references4 citations

TL;DR

This paper proposes a novel paradigm—Distributed Machine Intelligence (D-MI)—that decouples intelligence from applications by introducing a standardized intelligence layer between the device stack and application layer, enabling interoperable, reusable MI services across IoT devices. The approach enhances scalability, reduces latency, and boosts provider incentives by abstracting MI capabilities from specific use cases.

ABSTRACT

Machine Intelligence (MI) technologies have revolutionized the design and applications of computational intelligence systems, by introducing remarkable scientific and technological enhancements across domains. MI can improve Internet of Things (IoT) in several ways, such as optimizing the management of large volumes of data or improving automation and transmission in large-scale IoT deployments. When considering MI in the IoT context, MI services deployment must account for the latency demands and network bandwidth requirements. To this extent, moving the intelligence towards the IoT end-device aims to address such requirements and introduces the notion of Distributed MI (D-MI) also in the IoT context. However, current D-MI deployments are limited by the lack of MI interoperability. Currently, the intelligence is tightly bound to the application that exploits it, limiting the provisioning of that specific intelligence service to additional applications. The objective of this article is to propose a novel approach to cope with such constraints. It focuses on decoupling the intelligence from the application by revising the traditional device's stack and introducing an intelligence layer that provides services to the overlying application layer. This paradigm aims to provide final users with more control and accessibility of intelligence services by boosting providers' incentives to develop solutions that could theoretically reach any device. Based on the definition of this emerging paradigm, we explore several aspects related to the intelligence distribution and its impact in the whole MI ecosystem.

Motivation & Objective

• To address the lack of interoperability in current Machine Intelligence (MI) deployments within IoT systems.
• To reduce dependency of MI services on specific applications, enabling reuse across diverse devices and use cases.
• To improve latency and bandwidth efficiency in large-scale IoT deployments through edge-integrated intelligence.
• To empower end-users and providers with greater control and accessibility to MI services.
• To establish a foundational paradigm for a scalable, extensible MI ecosystem in IoT.

Proposed method

• Introduces a new intelligence layer that sits between the device stack and application layer, abstracting MI services from applications.
• Revises the traditional device architecture to support modular, reusable intelligence components.
• Designs the intelligence layer to expose standardized APIs for MI services, promoting interoperability.
• Enables dynamic discovery and composition of MI services across heterogeneous devices and platforms.
• Leverages edge computing to minimize latency and bandwidth usage in real-time IoT applications.
• Supports a microservices-like model for MI, allowing independent deployment and evolution of intelligence components.

Experimental results

Research questions

• RQ1 How can Machine Intelligence be decoupled from applications to enable reuse across diverse IoT devices?
• RQ2 What architectural changes are required to support interoperable, distributed MI in IoT systems?
• RQ3 How does the proposed intelligence layer improve latency and bandwidth efficiency in large-scale IoT deployments?
• RQ4 What incentives does the new paradigm create for providers to develop and share MI services?
• RQ5 How does the decoupling of intelligence from applications enhance user control and system extensibility?

Key findings

• The proposed intelligence layer enables MI services to be reused across multiple applications and devices, significantly improving interoperability.
• Decoupling intelligence from applications reduces vendor lock-in and increases flexibility in system design.
• The architecture reduces end-to-end latency by processing intelligence closer to data sources, aligning with edge computing principles.
• Standardized APIs for MI services enhance developer productivity and promote ecosystem growth.
• The model supports dynamic service composition, allowing applications to discover and integrate MI capabilities on demand.
• The approach creates stronger incentives for providers to innovate in MI, as services become platform-agnostic and reusable.