Research map / Architectures and efficiency / RNNs and normalization
Symplectic Adjoint Method for Exact Gradient of Neural ODE with Minimal Memory
Takashi Matsubara, Yuto Miyatake, Takaharu Yaguchi
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 |
|---|---|---|---|
| Use the symplectic adjoint method instead of the adjoint method | Neural ODE training · MiniBooNE | peak memory (MiB): 170 -> 95 | Improved |
| Use the symplectic adjoint method instead of the adjoint method | Neural ODE training · MiniBooNE | time (s/itr): 0.74 -> 0.84 | Hurt |
| Use the symplectic adjoint method instead of ACA | Neural ODE training · GAS | peak memory (MiB): 73 -> 20 | Improved |
| Use the symplectic adjoint method instead of ACA | Neural ODE training · GAS | time (s/itr): 3.98 -> 4.39 | Hurt |
| Use the symplectic adjoint method instead of ACA | Neural ODE training · HEPMASS | peak memory (MiB): 88 -> 35 | Improved |
| Use the symplectic adjoint method instead of ACA | Neural ODE training · HEPMASS | time (s/itr): 3.67 -> 4.15 | Hurt |
| Use the symplectic adjoint method instead of ACA | Neural ODE training · BSDS300 | peak memory (MiB): 757 -> 283 | Improved |
| Use the symplectic adjoint method instead of ACA | Neural ODE training · BSDS300 | time (s/itr): 6.97 -> 8.07 | Hurt |
| Use the symplectic adjoint method instead of the adjoint method | Neural ODE training · MNIST | time (s/itr): 10.12 -> 9.42 | Improved |
| Use the symplectic adjoint method instead of ACA | Neural ODE training · MNIST | peak memory (MiB): 4,332 -> 1,079 | Improved |
| Use the symplectic adjoint method instead of ACA | Neural ODE training · MNIST | time (s/itr): 7.94 -> 9.42 | Hurt |
| Use the symplectic adjoint method rather than ordinary backpropagation | Continuous normalizing flows · GAS | NLL: -9.53 -> -10.89 | Improved |
Closest papers by meaning
- "Hey, that’s not an ODE": Faster ODE Adjoints via Seminorms (ICML 2021 · similarity 0.88)
- AdjointDEIS: Efficient Gradients for Diffusion Models (NeurIPS 2024 · similarity 0.84)
- Global convergence of adjoint-optimized neural PDEs (NeurIPS 2026 · similarity 0.84)
- Efficient Training of Neural Fractional-Order Differential Equation via Adjoint Backpropagation (AAAI 2025 · similarity 0.84)
- AdjointDPM: Adjoint Sensitivity Method for Gradient Backpropagation of Diffusion Probabilistic Models (ICLR 2024 · similarity 0.83)
- How Deep Do We Need: Accelerating Training and Inference of Neural ODEs via Control Perspective (ICML 2024 · similarity 0.82)
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.
