Enhancing AI System Resiliency: Formulation and Guarantee for LSTM Resilience Based on Control Theory

2025/05/23 by Sota Yoshihara, Ryosuke Yamamoto, Yoshihara, Sota +5
Engineering · Health Professions · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #FOS: Electrical engineering #Fault Detection and Control Systems #Occupational Health and Safety Research #Systems and Control (eess.SY) #electronic engineering #information engineering

paper · pdf · doi:10.48550/arxiv.2505.17696

openalex publication_date 2025/05/23 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

This paper proposes a novel theoretical framework for guaranteeing and evaluating the resilience of long short-term memory (LSTM) networks in control systems. We introduce "recovery time" as a new metric of resilience in order to quantify the time required for an LSTM to return to its normal state after anomalous inputs. By mathematically refining incremental input-to-state stability (δISS) theory for LSTM, we derive a practical data-independent upper bound on recovery time. This upper bound gives us resilience-aware training. Experimental validation on simple models demonstrates the effectiveness of our resilience estimation and control methods, enhancing a foundation for rigorous quality assurance in safety-critical AI applications.

Citations

• Deep Long-Short Term Memory networks: Stability properties and Experimental validation
• Safe Nonlinear Control Using Robust Neural Lyapunov-Barrier Functions
• Reduced order modeling of fluid flows: Machine learning, Kolmogorov barrier, closure modeling, and partitioning
• LSTM Neural Networks: Input to State Stability and Probabilistic Safety Verification
• Learning of Multi-Context Models for Autonomous Underwater Vehicles
• How Robust are Deep Neural Networks?
• Long short-term memory neural network for traffic speed prediction using remote microwave sensor data
• Finite-Time Stability of Continuous Autonomous Systems

Related

• LSTM Neural Networks: Input to State Stability and Probabilistic Safety Verification
• Predicting Resilience with Neural Networks
• Measuring LTI System Resilience against Adversarial Disturbances based on Efficient Generalized Eigenvalue Computations
• Neural Network Adaptive Control with Long Short-Term Memory
• Quantitative Resilience of Linear Systems