SG-PALM: a Fast Physically Interpretable Tensor Graphical Model

Authors: Yu Wang, Alfred Hero

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the scalability and accuracy of SG-PALM for an important but challenging climate prediction problem: spatio-temporal forecasting of solar flares from multimodal imaging data.
Researcher Affiliation Academia Yu Wang 1 Alfred Hero 1 1University of Michigan, Ann Arbor, Michigan, USA.
Pseudocode Yes Algorithm 1 SG-PALM
Open Source Code Yes Both SG-PALM and Sy Glasso were implemented in Julia v1.5 (https: //github.com/ywa136/sg-palm).
Open Datasets Yes In this work, we illustrate the viability of the SG-PALM algorithm for solar flare prediction using data acquired by multiple instruments: the Solar Dynamics Observatory (SDO)/Helioseismic and Magnetic Imager (HMI) and SDO/Atmospheric Imaging Assembly (AIA). It is evident that these data contain information about the physical processes that govern solar activities (see Appendix E for detailed data descriptions). Additionally, the reference 'Galvez, R., Fouhey, D. F., Jin, M., Szenicer, A., Mu noz Jaramillo, A., Cheung, M. C., Wright, P. J., Bobra, M. G., Liu, Y., Mason, J., et al. A machine-learning data set prepared from the nasa solar dynamics observatory mission. The Astrophysical Journal Supplement Series, 242(1):7, 2019.' provides a formal citation to a publicly available dataset.
Dataset Splits No The paper mentions a 'training set' and 'testing samples' for the solar flare data but does not explicitly provide details about a separate validation set or specific train/validation/test split percentages in the main text. While regularization parameters were optimized on the training set, a distinct validation split is not described.
Hardware Specification Yes Experiments in this section were performed in a system with 8-core Intel Xeon CPU E5-2687W v2 3.40GHz equipped with 64GB RAM.
Software Dependencies Yes Both SG-PALM and Sy Glasso were implemented in Julia v1.5 (https: //github.com/ywa136/sg-palm).
Experiment Setup Yes The SG-PALM estimator was implemented using a regularization parameter λN = C1 q min(dk) log(d) N for all k with the constant C1 chosen by optimizing the prediction NRMSE on the training set over a range of λ values parameterized by C1. The data samples are summarized in d1 d2 d3 d4 tensors with q = d1 d2 d3 = 50 100 7 = 35000 and p = d4 = 13. The task is to predict the pth frame using the frames in each of the p 1 previous hours (i.e., one hour ahead prediction).