Research map / Time series / Causal and generative time series
MALI: A memory efficient and reverse accurate integrator for Neural ODEs
Juntang Zhuang, Nicha C Dvornek, sekhar tatikonda, James s Duncan
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 |
|---|---|---|---|
| Use MALI to train the Neural ODE, then evaluate with RK4 | Neural ODE18 · ImageNet validation | top-1 accuracy: 70.09 -> 70.21 | Improved |
| Replace adjoint training with MALI | Neural ODE18 · ImageNet validation | top-1 accuracy: 63% -> 70% | Improved |
| Replace ACA training with MALI | Latent-ODE · Mujoco test, 10% training data | MSE: 0.31 -> 0.35 | Hurt |
| Replace ACA training with MALI | Latent-ODE · Mujoco test, 20% training data | MSE: 0.27 -> 0.27 | No clear effect |
| Replace ACA training with MALI | Neural CDE · Speech Commands | accuracy: 93.2±0.2 -> 93.7±0.3 | Improved |
| Train FFJORD with MALI instead of SemiNorm | FFJORD · MNIST validation | bits per dimension: 0.96 -> 0.87 | Improved |
| Train FFJORD with MALI instead of Flow++ | FFJORD · CIFAR10 validation | bits per dimension: 3.28 -> 3.27 | Improved |
| Train FFJORD with MALI instead of Residual Flow | FFJORD · ImageNet64 validation | bits per dimension: 3.76 -> 3.71 | Improved |
| Set damping coefficient η to 0.95 instead of 1.0 (ablation) | Neural CDE · Speech Commands | accuracy: 93.7 ± 0.3 -> 93.7 ± 0.1 | No clear effect |
| Set damping coefficient η to 0.9 instead of 1.0 (ablation) | Neural CDE · Speech Commands | accuracy: 93.7 ± 0.3 -> 93.5 ± 0.2 | No clear effect |
| Set damping coefficient η to 0.85 instead of 1.0 (ablation) | Neural CDE · Speech Commands | accuracy: 93.7 ± 0.3 -> 93.7 ± 0.3 | No clear effect |
| Set damping coefficient η to 0.95 instead of 1.0 (ablation) | Latent-ODE · Mujoco test, 10% training data | MSE: 0.35 -> 0.36 | Hurt |
Closest papers by meaning
- Adaptive Checkpoint Adjoint Method for Gradient Estimation in Neural ODE (ICML 2020 · similarity 0.88)
- Improving Neural ODE Training with Temporal Adaptive Batch Normalization (NeurIPS 2024 · similarity 0.84)
- Simulation-Free Training of Neural ODEs on Paired Data (NeurIPS 2024 · similarity 0.84)
- DeltaDEQ: Exploiting Heterogeneous Convergence for Accelerating Deep Equilibrium Iterations (NeurIPS 2024 · similarity 0.83)
- How Deep Do We Need: Accelerating Training and Inference of Neural ODEs via Control Perspective (ICML 2024 · similarity 0.83)
- Learning to Solve Generative ODEs Beyond the Linear Span (NeurIPS 2026 · similarity 0.83)
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.
