Shape analysis for time series

Authors: Thibaut Germain, Samuel Gruffaz, Charles Truong, Alain Durmus, Laurent Oudre

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

Reproducibility Variable Result LLM Response
Research Type Experimental We showcase the advantages of our representation compared to existing methods using synthetic data and real-world examples motivated by biomedical applications.
Researcher Affiliation Academia Thibaut Germain1 Centre Borelli, ENS Paris-Saclay 4 av. des sciences, 91190 Samuel Gruffaz1 Centre Borelli, ENS Paris-Saclay 4 av. des sciences, 91190 Charles Truong1 Centre Borelli, ENS Paris-Saclay 4 av. des sciences, 91190 Laurent Oudre1 Centre Borelli, ENS Paris-Saclay 4 av. des sciences, 91190 Alain Durmus CMAP, CNRS, Ecole polytechnique Institut Polytechnique de Paris 91120 Palaiseau, France
Pseudocode No The paper describes algorithms and equations but does not provide a formal pseudocode block or algorithm environment labeled as such.
Open Source Code Yes The source code is available on Github4. https://github.com/thibaut-germain/TSLDDMM
Open Datasets Yes We selected 15 shape-based datasets (7 univariates and 8 multivariates) from the from the University of East Anglia (UEA) and the University of California Riverside (UCR) Time Series Classification Repository8 [15, 3].
Dataset Splits Yes Spilt the dataset in train 75%, validation 15%, and test 15%.
Hardware Specification Yes All experiments were performed on a Debian 6.1.69-1 server with NVIDIA RTX A2000 12GB GPU, Intel(R) Xeon(R) Gold 5220R CPU @ 2.20GHz, and 250 GB of RAM.
Software Dependencies No The paper mentions specific libraries like "JAX library" and "OPTAX library" but does not include specific version numbers for these or other key software components, which is required for reproducibility.
Experiment Setup Yes The optimization hyperparameter details are given in Appendix E.1. By default, we set nb_steps to 400 and ηM to 0.1. To learn TS-LDDMM (resp. LDDMM) representations, the velocity field kernel KG is set to (c0, c1, σT,0, σT,1, σx) = (1, 0.1, 0.33 l, 1, nd), (resp. (σT , σx) = (0.33 l, nd)) where l is the average time series length and nd the number of dimensions.