Physics-Informed Neural Network Policy Iteration: Algorithms, Convergence, and Verification

Authors: Yiming Meng, Ruikun Zhou, Amartya Mukherjee, Maxwell Fitzsimmons, Christopher Song, Jun Liu

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present numerical examples to evaluate the performance of the proposed algorithms.
Researcher Affiliation Academia 1Coordinated Science Laboratory, University of Illinois Urbana-Champaign, Champaign Urbana, Illinois, United States 2Department of Applied Mathematics, University of Waterloo, Waterloo, Canada.
Pseudocode Yes Algorithm 1 Extreme Learning Machine Policy Iteration (ELM-PI) and Algorithm 2 Physics-Informed Neural Network Policy Iteration (PINN-PI)
Open Source Code Yes The code of these experiments can be found at https://git.uwaterloo. ca/hybrid-systems-lab/lyznet/.
Open Datasets No The paper uses synthetic and benchmark control problems defined by their dynamics, but does not refer to publicly available datasets with access information.
Dataset Splits No The paper mentions training and testing errors and the number of collocation points, but does not specify explicit training/validation/test dataset splits with percentages or counts.
Hardware Specification Yes ELM-PI experiments were run with an Intel Gold 6148 Skylake @ 2.4 GHz, and PINN-PI experiments were run with an NVidia V100SXM2 (16G memory).
Software Dependencies No The paper mentions software like 'Adam', 'd Real', and 'stable-baselines3' but does not provide specific version numbers for these dependencies.
Experiment Setup Yes All instances of ELM-PI and PINN-PI are run with the tanh activation function, unless otherwise noted. ... For PINN-PI, we train the network for 10,000 steps in each iteration with Adam. ... The iteration number of PI is set to be 10.