[Paper Review] TinyML for Ubiquitous Edge AI

Stanislava Soro|arXiv (Cornell University)|Feb 2, 2021

Advanced Neural Network Applications13 references33 citations

TL;DR

TinyML enables deep learning on ultra-low-power embedded devices, enabling distributed edge inference and autonomous reasoning without heavy cloud reliance; the paper surveys challenges and technological enablers.

ABSTRACT

TinyML is a fast-growing multidisciplinary field at the intersection of machine learning, hardware, and software, that focuses on enabling deep learning algorithms on embedded (microcontroller powered) devices operating at extremely low power range (mW range and below). TinyML addresses the challenges in designing power-efficient, compact deep neural network models, supporting software framework, and embedded hardware that will enable a wide range of customized, ubiquitous inference applications on battery-operated, resource-constrained devices. In this report, we discuss the major challenges and technological enablers that direct this field's expansion. TinyML will open the door to the new types of edge services and applications that do not rely on cloud processing but thrive on distributed edge inference and autonomous reasoning.

Motivation & Objective

• Identify the main challenges facing TinyML on embedded devices.
• Outline the technological enablers (hardware, software, and model design) that enable TinyML.
• Explain how TinyML can enable edge services and autonomous reasoning without cloud dependence.

Proposed method

• Provide a conceptual overview of TinyML and its convergence of ML, hardware, and software.
• Discuss major challenges in power, size, and efficiency for embedded DNNs.
• Describe technological enablers including hardware, software frameworks, and model design.
• Discuss potential edge inference scenarios and autonomous capabilities enabled by TinyML.

Experimental results

Research questions

• RQ1 What are the key challenges and technological enablers that will drive the expansion of TinyML?
• RQ2 How can TinyML enable ubiquitous edge inference and autonomous edge reasoning without cloud processing?
• RQ3 What combination of hardware, software, and model design is required to support power-efficient, compact DNNs on microcontrollers?

Key findings

• TinyML is poised to enable new edge services that operate with distributed inference and autonomous reasoning rather than relying on cloud processing.
• Power-efficient, compact deep neural networks are central to enabling ubiquitous edge AI on battery-operated devices.
• Advancements in embedded hardware and software frameworks are essential enablers for practical TinyML deployment.