analiza aiqbench
The paper argues that a neural network trained offline on one state distribution can shift the system into different states once it is deployed in a closed-loop controller, which degrades performance. It offers a taxonomy of the resulting errors and proposes time-scale-separated online adaptation, validated only on a nonlinear Duffing oscillator in simulation. For ML practitioners working on safety-critical control, the practical point is that static offline training may not hold up after deployment, but the evidence so far is limited to a single toy system.
Learning Changes the World and Then What Was Learned has Changed: Training and Deploying Neural Networks in the Control Loop
arXiv:2610.22084v1 Announce Type: cross Abstract: Neural networks offer critical adaptability for controlling complex systems where nominal modeling proves intractable. However, when offline-trained neural networks are deployed within closed-loop control systems, their actions fundamentally alter the state distribution upon which they were trained. This paper introduces a unified taxonomy to categorize these sequential failure modes: the initial Closed-Loop Identification Error, the deployment-induced Action Error, and the continuous Tracking Error. Through discrete-time sensitivity analysis, this work formalizes how offline-learned parameters locally guarantee a trajectory divergence upon deployment, degrading performance due to local loss geometry. To mitigate this distribution shift, we propose actionable design policies centered on time-scale separated online adaptation. These mechanisms are validated through ablation studies on a nonlinear Duffing oscillator, demonstrating that bounded continuous learning successfully mitigates trajectory divergence whereas static deployments fail, offering concrete guidelines for practitioners deploying machine learning components in safety-critical feedback systems.
Abstract:Neural networks offer critical adaptability for controlling complex systems where nominal modeling proves intractable. However, when offline-trained neural networks are deployed within closed-loop control systems, their actions fundamentally alter the state distribution upon which they were trained. This paper introduces a unified taxonomy to categorize these sequential failure modes: the initial Closed-Loop Identification Error, the deployment-induced Action Error, and the continuous Tracking Error. Through discrete-time sensitivity analysis, this work formalizes how offline-learned parameters locally guarantee a trajectory divergence upon deployment, degrading performance due to local loss geometry. To mitigate this distribution shift, we propose actionable design policies centered on time-scale separated online adaptation. These mechanisms are validated through ablation studies on a nonlinear Duffing oscillator, demonstrating that bounded continuous learning successfully mitigates trajectory divergence whereas static deployments fail, offering concrete guidelines for practitioners deploying machine learning components in safety-critical feedback systems.
Machine Learning (stat.ML); Machine Learning (cs.LG)
arXiv:2610.22084 [stat.ML]
(or arXiv:2610.22084v1 [stat.ML] for this version)
https://doi.org/10.48550/arXiv.2610.22084 arXiv-issued DOI via DataCite (pending registration)
(Lippi, 2026) In Proc. of the 23rd Int. Conf. on Informatics in Control, Automation and Robotics - Vol. 1, ISBN 978-989-758-870-9, ISSN 2184-2809, pp 15-26
Text integral sub licența deschisă a publicației · 187 cuvinte
Subiecte conexe
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · 3 aug. 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · 25 sept. 2026
Echoverse: Deep, evolving environments for computer-use agents
Microsoft Research · 30 iul. 2026
Game Design Creativity evals
LessWrong · 12 oct. 2026