Research map / Probabilistic and scientific modeling / Variational and generative sampling
Interpolation Technique to Speed Up Gradients Propagation in Neural ODEs
Talgat Daulbaev, Alexandr Katrutsa, Larisa Markeeva, Julia Gusak, Andrzej Cichocki, Ivan Oseledets
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace reverse dynamic method with interpolated activations (IRDM, N=16) | neural ODE training · toy density estimation (circles) | training time (s): 7874 -> 6842 | Improved |
| replace reverse dynamic method with IRDM (N=16) | neural ODE training · toy density estimation (2spirals) | training time (s): 9285 -> 8938 | Improved |
| replace RDM with IRDM | neural ODE training · miniboone density estimation | test loss vs training time | Improved |
| replace RDM with IRDM | neural ODE training · toy dynamical system reconstruction | right-hand side evaluations: at least 10 times fewer | Improved |
| replace RDM with IRDM | variational autoencoder · caltech, freyfaces | final test loss and training time | No clear effect |
| replace RDM with IRDM | neural ODE · CIFAR10 classification | test accuracy and training time: 0.867 test accuracy | Improved |
| use too few Chebyshev nodes (N=4) instead of more (ablation) | IRDM · toy density estimation | test loss | Hurt |
| use too many Chebyshev nodes (N=128) instead of N=16 (ablation) | IRDM · pinwheel density estimation | training time (s): 4750 -> 5574 | Hurt |
| use too many Chebyshev nodes (N=128) instead of N=16 (ablation) | IRDM · circles density estimation | training time (s): 6842 -> 8113 | Hurt |
| use barycentric Lagrange interpolation instead of piecewise-linear interpolation (ablation) | IRDM · MNIST classification | test accuracy stability | Improved |
| decrease DOPRI5 tolerance (ablation) | IRDM · toy 2D dynamical system | gradient error | Improved |
| increase number of Chebyshev nodes (ablation) | IRDM · toy 2D dynamical system | gradient approximation quality | Hurt |
Closest papers by meaning
- How to Train Your Neural ODE: the World of Jacobian and Kinetic Regularization (ICML 2020 · similarity 0.90)
- "Hey, that’s not an ODE": Faster ODE Adjoints via Seminorms (ICML 2021 · similarity 0.89)
- How Deep Do We Need: Accelerating Training and Inference of Neural ODEs via Control Perspective (ICML 2024 · similarity 0.89)
- Do Residual Neural Networks discretize Neural Ordinary Differential Equations? (NeurIPS 2022 · similarity 0.88)
- Global Convergence in Neural ODEs: Impact of Activation Functions (ICLR 2025 · similarity 0.88)
- Simulation-Free Training of Neural ODEs on Paired Data (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
