FINDE: Neural Differential Equations for Finding and Preserving Invariant Quantities

Authors: Takashi Matsubara, Takaharu Yaguchi

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluated FINDE and base models using datasets associated with first integrals; these are summarized in Table 2. A gravitational two-body problem (2-body) on a 2-dimensional configuration space is a typical Hamiltonian system in the canonical form. In addition to the total energy, the system has first integrals related to symmetries in space, namely, the linear and angular momenta.
Researcher Affiliation Academia Takashi Matsubara Osaka University Toyonaka, Osaka, 560 8531 Japan matsubara@sys.es.osaka-u.ac.jp Takaharu Yaguchi Kobe University Kobe, Hyogo, 657 8501 Japan yaguchi@pearl.kobe-u.ac.jp
Pseudocode No The paper does not contain pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes We implemented all codes by modifying the officially released codes of HNN (Greydanus et al., 2019) 1 and DGNet (Matsubara et al., 2020)2. ... The authors have enclosed the source code for generating the datasets and running the experiments as supplementary material.
Open Datasets Yes We generated a time-series set of each dataset with different initial conditions (hence, different values of first integrals). See Appendix C for more details. ... The authors have enclosed the source code for generating the datasets and running the experiments as supplementary material.
Dataset Splits No The paper specifies the number of time-series for 'training' and 'evaluation' (testing) but does not mention specific percentages or counts for a separate validation split.
Hardware Specification Yes All experiments were performed on a single NVIDIA A100.
Software Dependencies Yes We used Python v. 3.8.12 with packages scipy v. 1.7.3, pytorch v. 1.10.2, torchdiffeq v. 0.1.1, functorch v. 1.10 preview, and gplearn v. 0.4.2.
Experiment Setup Yes We used fully-connected neural networks with two hidden layers. The input was the state u, and the output represented the first integrals V for FINDE, time-derivative ˆf for NODE, or the Hamiltonian H for HNN. Each hidden layer had 200 units and preceded a hyperbolic tangent activation function. Each weight matrix was initialized as an orthogonal matrix. ... The base model and FINDE were jointly trained using the Adam optimizer (Kingma & Ba, 2015) with the parameters (β1, β2) = (0.9, 0.999) and a batch size of 200. The learning rate was initialized to 10^-3 and decayed to zero with cosine annealing (Loshchilov & Hutter, 2017).