Consistent Long-Term Forecasting of Ergodic Dynamical Systems

Authors: Vladimir R Kostic, Karim Lounici, Prune Inzerilli, Pietro Novelli, Massimiliano Pontil

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

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments illustrate the advantages of our approach in practice. [...] In Section 7 we present numerical experiments with our approach.
Researcher Affiliation Academia 1Italian Institute of Technology, Genoa, Italy 2University of Novi Sad, Serbia 3CMAP, Ecole Polytechnique, Palaiseau, France 4University College London, UK.
Pseudocode Yes Algorithm 1 Forecasting observables and measures with KRR/PCR/RRR estimator via DLI framework
Open Source Code No The code to reproduce the experiments will be open sourced.
Open Datasets Yes Angles of Alanine Dipeptide We assess the forecasting performance of DLI estimators on a dataset of molecular dynamics simulations for the small molecule Alanine Dipeptide (Wehmeyer & No e, 2018).
Dataset Splits Yes Each experiment has been repeated 100 times independently, and the hyperparameters were tuned on a validation set of 500 points sampled from the invariant distribution.
Hardware Specification Yes The experiments were run on a workstation equipped with an Intel(R) Core i9-9900X CPU @ 3.50GHz, 48GB of RAM and a NVIDIA Ge Force RTX 2080 Ti GPU.
Software Dependencies No All experiments have been implemented in Python 3.11. For the Cox Ingersoll Ross, the Reduced Rank Regression estimator was implemented using the reference code from (Kostic et al., 2022) available at https://github.com/Machine-Learning-Dynamical-Systems/kooplearn.
Experiment Setup No The paper states that 'hyperparameters were tuned on a validation set', but it does not provide the specific values for these hyperparameters (e.g., learning rate, batch size, optimizer settings) or other concrete experimental setup details in the main text.