[ONLINE] Evolutionary Optimisation of Neural Network Hardware Accelerators

Monday 3 Nov 2025, 14:00 → 15:45 Europe/Prague

ZOOM (ONLINE)

Annotation

Neural networks are not just considered the domain of large GPUs and supercomputers. Increasingly, they are also found in simple embedded systems. However, these systems are limited in terms of computational and memory resources. In this tutorial, methods based on evolutionary algorithms for optimising both accelerators for specific networks and networks for specific accelerators will be shown. It will be demonstrated how the targeted introduction of error into computation, improving the organisation of computational units and memory, can affect the overall efficiency of inference. Using Capsule Networks as an example, the modification of the network architecture to improve parameter efficiency in hardware will be demonstrated, and similarly, small Ternary Networks will be optimised for printed electronics.

Benefits for the attendees, what will they learn

Participants will gain an overview of the category of smaller networks used in embedded systems and advanced optimisation techniques. They will learn about power estimation techniques, mapping issues to compute units and the concept of approximate computing.

Level

Beginner

Language

English

Tutor

Ing. Vojtěch Mrázek, Ph.D.

Ing. Vojtěch Mrázek, Ph.D. received a M.Sc. and Ph.D. degrees in information technology from the Faculty of Information Technology, Brno University of Technology, Czech Republic, in 2014 and 2018. He is a researcher at the Faculty of Information Technology with Evolvable Hardware Group and he was also a visiting post-doc researcher at Institute of Computer Engineering, Technische Universität Wien (TU Wien), Vienna, Austria (2018-2019). His research interests are approximate computing, genetic programming and machine learning. He has authored or co-authored over 60 conference/journal papers focused on approximate computing and evolvable hardware. He received several awards for his research in approximate computing, including the Joseph Fourier Award in 2018 for research in computer science and engineering and Czech Mind – Doctorandus for outstanding research work in the technical field for PhD student in 2019.

Acknowledgements

This work has been supported by Czech Science Foundation project GA24-10990S

LUMI AI Factory is funded jointly by the EuroHPC Joint Undertaking, through the European Union's Connecting Europe Facility and the Horizon 2020 research and innovation programme, as well as Finland, the Czech Republic, Poland, Estonia, Norway, and Denmark.

 

This course was supported by the Ministry of Education, Youth and Sports of the Czech Republic through the e-INFRA CZ (ID:90254). 

 

All presentations and educational materials of this course are provided under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.

Evolutionary Optimisation Presentation.pdf

Video Evolutionary Optimisation of Neural Network Hardware Accelerators

Presentation

Overview of the workshop objectives.

Discussion

Integrating native code in Python enables developers to leverage high-performance libraries or custom compiled code for faster execution of computationally intensive tasks.