Published in AAAI-26, 2026
LILAD is a new in-context learning framework for system identification that jointly ensures adaptability and Lyapunov stability, even under distribution shifts. Trained across diverse tasks, it produces an adaptive dynamics model and adaptive Lyapunov certificate that rapidly specialize to new system instances under parametric uncertainty, using only a short trajectory prompt.
Download here
Published in TSG, 2025
This paper introduces an LLM-based adaptive inverter control scheme for distribution voltage regulation under frequent topology changes. By using in-context learning to build a topology-adaptive power flow surrogate and integrating it with LSTM-based load forecasting and MPC, the method adapts to both known and unseen topologies with minimal fine-tuning. Its effectiveness is demonstrated on a modified IEEE 123-bus test system.
Download here
Published in AAAI-24, 2024
This paper proposes a meta-learning-based adaptive Lyapunov function that efficiently adapts to any test-time instance of a dynamical system with parametric uncertainty. As a result, a single Lyapunov function can be trained and then quickly fine-tuned at test time to assess the stability of any arbitrary system instance.
Download here
Published in CDC-21, 2021
We develop a distributed learning framework for assessing the stability of networked nonlinear systems by combining neural network–based storage functions, SMT-based dissipativity verification, and an ADMM-based global stability check. The method yields a network-level Lyapunov function and region of attraction, demonstrated on a microgrid interconnection.
Download here