Differentiable Multiple Shooting Layers

Authors: Stefano Massaroli, Michael Poli, Sho Sonoda, Taiji Suzuki, Jinkyoo Park, Atsushi Yamashita, Hajime Asama

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

Reproducibility Variable Result LLM Response
Research Type Experimental MSLs are showcased in long horizon optimal control of ODEs and PDEs and as latent models for sequence generation. Finally, we investigate the speedups obtained through application of MSL inference in neural controlled differential equations (Neural CDEs) for time series classification of medical data.
Researcher Affiliation Academia Stefano Massaroli The University of Tokyo, Diff Eq ML massaroli@robot.t.u-tokyo.ac.jp Michael Poli KAIST, Diff Eq ML poli_m@kaist.ac.kr Sho Sonoda RIKEN Taiji Suzuki The University of Tokyo, RIKEN Jinkyoo Park KAIST Atsushi Yamashita The University of Tokyo Hajime Asama The University of Tokyo
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper states in a checklist 'Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]', but the main text does not provide a specific URL or an explicit statement about the release of their own source code.
Open Datasets Yes We tackle the Physio Net 2019 challenge (Goldberger et al., 2000) on sepsis prediction, following the exact experimental procedure described by Kidger et al. (2020b), including hyperparameters and Neural CDE architectures.
Dataset Splits No The paper states 'We train all models on the full dataset to enable application of the fixed point tracking technique for MSLs', which implies no explicit train/validation/test split for reproducibility in the main text.
Hardware Specification Yes We note that the training for all models has been performed on a single NVIDIA RTX A6000 with 48Gb of GPU memory.
Software Dependencies No The paper mentions 'dopri5 solver' and 'FENICS project version 1.5' but does not provide a comprehensive list of software dependencies with specific version numbers (e.g., Python, PyTorch/JAX, CUDA versions) for their own implementation.
Experiment Setup No The paper states 'The Neural ODE is solved via the dopri5 solver with absolute and relative tolerances set to 10 4' and 'following the exact experimental procedure described by Kidger et al. (2020b), including hyperparameters and Neural CDE architectures.' However, it defers to another paper for hyperparameters rather than listing them directly, and does not provide a comprehensive set of specific training configurations.