$S^3$ method optimizes neural network abstractions for dynamical systems

Researchers have developed a new method called a smooth simulation surrogate ($S^3$) to optimize discrete abstractions of dynamical systems, particularly those with neural network controllers. This differentiable objective approximates a metric used to quantify conservatism in abstractions, allowing for gradient-based optimization while maintaining soundness. Evaluations on three case studies demonstrated that $S^3$ correlates well with the reverse simulation metric, is computationally efficient, and effectively reduces abstraction conservatism. AI

IMPACT Introduces a novel optimization technique for improving the safety and reliability of neural network controllers in critical systems.

RANK_REASON Academic paper detailing a new method for optimizing abstractions of dynamical systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Bi-simulation theory
• Electrical Engineering and Systems Science
• Hugging Face
• Neural Networks
• $S^3$
• Systems and Control
• Taylor model-based reachability

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →