Talks and Presentations

On-Device Training of Fully Quantized Deep Neural Networks on Cortex-M Microcontrollers

July 03, 2024

Poster, Embedded Talk: Next Generation IoT - Kommunikation & KI, Nürnberg, Germany

This poster shows how fully quantized DNNs deployed on Cortex-M microcontroller units (MCUs) can be either fine-tuned or even retrained from scratch. DNN training on MCUs poses additional challenges due to the tight resource constraints of such systems, and this poster discusses a methodology to overcome them.
More information | Poster

Optimizing Neural Networks with Multi Objective Bayesian Optimization and Augmented Random Search

March 14, 2023

Poster, Workshop on Optimization and Machine Learning, Nürnberg and Waischenfeld, Germany

This poster explores how multiobjective Bayesian optimization for hardware-aware neural architecture search for microcontroller targets can be improved with reinforcement learning, in particular Augmented Random Search (ARS).
More information | Poster

Multi-Objective Bayesian Optimization of Deep Neural Networks for Deployment on Microcontrollers

March 02, 2023

Talk, 4th Workshop on Embedded Machine Learning - WEML2023, Heidelberg University, Germany

This presentation discusses efficient design and deployment of DNNs on microcontroller units (MCUs) using both pruning and quantization techniques. It also provides an overview of automated hardware-aware DNN design for MCU targets using multi-objective Bayesian optimization.
More information | Slides

Mark Deutel

Talks and Presentations

On-Device Training of Fully Quantized Deep Neural Networks on Cortex-M Microcontrollers

Optimizing Neural Networks with Multi Objective Bayesian Optimization and Augmented Random Search

Multi-Objective Bayesian Optimization of Deep Neural Networks for Deployment on Microcontrollers