Model compression for neural network controllers: a tutorial survey with a focus on controllers with latent state space
at - Automatisierungstechnik · 26.03.2026 · 74 (3) · 212-232
DOI: 10.1515/auto-2025-0107 ↗Kurz gefasst
Fachartikel in der Zeitschrift at - Automatisierungstechnik, online erschienen am 09.03.2026. Verfasst von Ganesh Sundaram und 2 weiteren Autor:innen. Die Arbeit nennt 84 Literaturangaben.
Abstract (Originalsprache)
Model compression is key for deploying Deep Neural Networks on resource-constrained hardware. Its application to Neural Network Controllers (NNCs) is challenging because it can compromise control-theoretic properties and performance, a critical issue for modern controllers that use latent-space models. This paper surveys and empirically evaluates compression techniques for these models, using an MNIST autoencoder and a Temporal Difference Model Predictive Control agent as test cases across diverse hardware. We find that general compression techniques apply to latent-space models and that careful compression can preserve the theoretical properties of NNCs. Specific findings indicate that quantization can increase latency on non-specialized hardware, fine-tuning is crucial for performance recovery, and hybrid methods yield the best trade-offs.
Zitieren
APA
Sundaram, G., Ulmen, J., & Görges, D. (2026). Model compression for neural network controllers: a tutorial survey with a focus on controllers with latent state space. at - Automatisierungstechnik, 74(3), 212-232. https://doi.org/10.1515/auto-2025-0107
Vancouver
Sundaram G, Ulmen J, Görges D. Model compression for neural network controllers: a tutorial survey with a focus on controllers with latent state space. at - Automatisierungstechnik. 2026;74(3):212-232. doi:10.1515/auto-2025-0107
BibTeX
@article{Sundaram2026,
author = {Sundaram, Ganesh and Ulmen, Jonas and Görges, Daniel},
title = {Model compression for neural network controllers: a tutorial survey with a focus on controllers with latent state space},
journal = {at - Automatisierungstechnik},
year = {2026},
volume = {74},
number = {3},
pages = {212--232},
doi = {10.1515/auto-2025-0107}
}
EUVIMED