Magics will present new research at SpAIce 2026, taking place from 21–23 October in Noordwijk, the Netherlands.
During the conference, Toon Vinck - Software Engineer at Magics, will present the paper: "The Effect of Structured Pruning on the Fault Tolerance of Convolutional Neural Networks in Radiation Environments."
As artificial intelligence becomes increasingly integrated into satellite payloads and onboard processing systems, reducing computational complexity has become an important design objective. Structured pruning is widely used to reduce model size and improve inference efficiency, but its impact on radiation-induced faults has remained largely unexplored.
AI efficiency versus reliability
The paper investigates how structured pruning influences the fault tolerance of convolutional neural networks operating in radiation environments.
A common assumption is that reducing network redundancy may increase sensitivity to radiation-induced errors. The study evaluates this trade-off and demonstrates that the reduction in execution time achieved through pruning can compensate for the increased fault sensitivity by reducing the overall exposure window to radiation-induced events.
The result is a promising approach for developing more computationally efficient onboard AI systems while maintaining overall system reliability.
As onboard AI continues to support Earth observation, autonomous spacecraft operations, and future exploration missions, understanding these design trade-offs becomes increasingly important for mission-critical electronics.
Meet us at SpAIce 2026
If you're attending SpAIce 2026, we'd be pleased to discuss reliable AI for space systems, radiation effects on machine learning, and the role of radiation-hardened electronics in future onboard computing architectures.
SpAIce 2026
📅 21–23 October 2026
📍 Noordwijk, The Netherlands
The paper is available as a preprint on arXiv:
http://arxiv.org/abs/2610.03117
