Stability-Informed Initialization of Neural Ordinary Differential Equations

Authors: Theodor Westny, Arman Mohammadi, Daniel Jung, Erik Frisk

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

Reproducibility Variable Result LLM Response
Research Type Experimental The effectiveness of the initialization method is displayed across several learning benchmarks and industrial applications. To assess the performance and usability of the proposed technique, several experiments are conducted across a diverse set of problem domains.
Researcher Affiliation Academia 1Department of Electrical Engineering, Link oping University, Link oping, Sweden.
Pseudocode Yes Algorithm 1 Stability Region Rejection Sampling
Open Source Code Yes Implementations are made publicly available.2 https://github.com/westny/neural-stability
Open Datasets Yes the original MNIST (Le Cun et al., 1998) and CIFAR10 (Krizhevsky et al., 2009) datasets are modified for a sequential classification task. The second dataset, referred to as the Human Activity dataset (Vidulin et al., 2010), features data from five individuals, each wearing four localization tags (left ankle, right ankle, belt, chest). The third dataset, referenced as the Air Quality dataset (Vito, 2016) encompasses a set of air quality measurements from an urban monitoring station in Italy.
Dataset Splits No The paper describes training and testing phases for different experiments but does not explicitly provide specific percentages or counts for training, validation, and test dataset splits.
Hardware Specification No Computations were enabled by the Berzelius resource provided by the Knut and Alice Wallenberg Foundation at the National Supercomputer Centre.
Software Dependencies No No specific software versions (e.g., Python 3.x, PyTorch 1.x) or versions of other key libraries/solvers were explicitly listed.
Experiment Setup No The paper mentions training with 'different hyperparameters' and that 'batch size and learning rate may differ with the task but is always the same for all methods', but it does not provide concrete values for these hyperparameters or other training settings like epochs or optimizer details in the main text.