Explore energy-efficient on-device processing for next-generation endpoint machine learning in this tinyML Summit 2020 presentation by Tomas Edso, Senior Principal Engineer at Arm. Dive into best-in-class solutions optimized for endpoint AI and learn about unified software development for the fastest path to implementation. Discover the Cortex-M55, the most AI-capable Cortex-M CPU, and its simplified software development based on a unified programmer's view. Examine Cortex-M55 and CMSIS-NN performance results, and gain insights into the Ethos-U55 overview, including typical data flow and interfaces. Understand the mapping of neural networks to hardware using TensorFlow Lite, and explore an example smart speaker pipeline with throughput considerations. Analyze a typical ML workload for a voice assistant and gain valuable knowledge on advancing endpoint machine learning capabilities.
Energy-efficient On-device Processing for Next-generation Endpoint ML - tinyML Summit 2020
Overview
Syllabus
Intro
Best-in-class Solution Optimized for Endpoint Al
Unified Software Development: Fastest Path to Endpoint Al
Cortex-M55:The most Al-capable Cortex-M CPU
Simplified Software Development Based on a Unified Programmer's View
Cortex-M55 and CMSIS-NN performance results
Ethos-U55 overview
Typical Ethos-U55 data flow
Ethos-U55 interfaces
Mapping of NNs to Hardware using TensorFlow Lite
An example smart speaker pipeline
Throughput-smart speaker use case
Example: Typical ML Workload for a Voice Assistant
Taught by
tinyML
