Research map / Graph learning / Graph algorithms and clustering
Adaptive Checkpoint Adjoint Method for Gradient Estimation in Neural ODE
Juntang Zhuang, Nicha Dvornek, Xiaoxiao Li, Sekhar Tatikonda, Xenophon Papademetris, James 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 |
|---|---|---|---|
| Train NODE18 with ACA instead of the adjoint method | NODE18 · CIFAR10 test | error rate: 4.92 -> 4.85 | Improved |
| Train NODE18 with ACA instead of the naive method | NODE18 · CIFAR10 test | error rate: 5.29 -> 4.85 | Improved |
| Use ACA-trained NODE18 rather than ResNet18 | image classification · CIFAR10 test | error rate: 9.8 (*19) -> 4.85 | Improved |
| Test ACA-trained NODE18 with alternate adaptive solvers (ablation) | NODE18-ACA · CIFAR10 and CIFAR100 test | error rate: CIFAR10: 4.85 -> 5.27, 5.24, 5.29; CIFAR100: 22.66 -> 24.44, 24.43, 24.43 | No clear effect |
| Train Latent-ODE with ACA instead of the adjoint method | Latent-ODE · Mujoco irregularly sampled time series | test MSE (×10−2): 0.47 -> 0.31 | Improved |
| Train Latent-ODE with ACA instead of the naive method | Latent-ODE · Mujoco irregularly sampled time series | test MSE (×10−2): *0.36 -> 0.31 | Improved |
| Add partial physical information through augmented inputs | LSTM · Three-body future trajectory | Test MSE: 0.59+-0.12 -> 0.49+-0.06 | Improved |
| Use NODE-ACA with partial physical information rather than LSTM-aug-input | three-body trajectory model · Three-body future trajectory | Test MSE: 0.49+-0.06 -> 0.16+-0.06 | Improved |
| Train the physical ODE model with ACA instead of the adjoint method | ODE model · Three-body future trajectory | Test MSE: 0.0025+-0.0012 -> 0.0007+-0.0005 | Improved |
| Train the physical ODE model with ACA instead of the naive method | ODE model · Three-body future trajectory | Test MSE: 0.0025+-0.0013 -> 0.0007+-0.0005 | Improved |
| Use ACA instead of adjoint gradient estimation | NODE18 · CIFAR100 test | error rate: 24.13 -> 22.66 | Improved |
Closest papers by meaning
- AdjointDPM: Adjoint Sensitivity Method for Gradient Backpropagation of Diffusion Probabilistic Models (ICLR 2024 · similarity 0.86)
- Efficient Adjoint Matching for Fine-tuning Diffusion Models (NeurIPS 2026 · similarity 0.84)
- Non-asymptotic Analysis of Biased Adaptive Stochastic Approximation (NeurIPS 2024 · similarity 0.84)
- AdjointDEIS: Efficient Gradients for Diffusion Models (NeurIPS 2024 · similarity 0.84)
- A Comprehensive Framework for Analyzing the Convergence of Adam: Bridging the Gap with SGD (ICML 2025 · similarity 0.83)
- Regularized Adaptive Momentum Dual Averaging with an Efficient Inexact Subproblem Solver for Training Structured Neural Network (NeurIPS 2024 · 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.
