Erik Lundell

| Kite - Bootstrap Admin Template

Speaker

Software Engineer
Arm
Joined February 2025

LIS25-323 Optimizing Edge AI with Executorch and Ethos-U

Technical presentation - 30 minutes (including q&a)

Ai machine learning

Friday, 16 May 12:40 - 13:05 (Europe/Lisbon)
Your timezone: Friday, 16 May 12:40 - 13:05 (Europe/Lisbon)
Session room 3 | Opala III

In this presentation, we explore the integration of Arm's Ethos-U NPUs with Executorch to enable efficient edge AI solutions. We leverage the TOSA (Tensor Operator Set Architecture) framework to optimize neural network operations, ensuring seamless compatibility and performance across various hardware platforms. By utilizing PyTorch and Executorch, we demonstrate how to deploy and run PyTorch models on edge devices, enhancing AI capabilities with minimal latency and power consumption. Will will also talk about how we integrated the Corstone Fixed Virtual Platform (FVP) provides a robust simulation environment, allowing for comprehensive testing and validation of our edge AI solutions. This integration not only accelerates development but also ensures scalable and efficient deployment of AI models on edge devices.

About

Erik Lundell is a Software Engineer at Arm, specializing in the development of Executorch. With a strong background in machine learning and edge computing, Erik has been instrumental in integrating advanced AI capabilities into edge devices. His work focuses on leveraging frameworks like TOSA and PyTorch to enhance the performance and efficiency of neural network operations on Arm's Ethos-U NPUs. Passionate about pushing the boundaries of technology, Erik is dedicated to enabling innovative and efficient AI deployments at the edge.