End-to-End Learning Framework for Solving Non-Markovian Optimal Control
Published in ICML 2025, 2025
This work develops the first end-to-end learning framework (FOLOC) for fractional-order linear time-invariant (FOLTI) systems that integrates system identification and optimal control derivation under non-Markovian dynamics. The framework is theoretically founded—extending LQR to fractional-order settings and providing sample complexity guarantees—and validated empirically in noisy, realistic environments. Code is publicly available.