Research map / Probabilistic and scientific modeling / Variational and generative sampling
Hypersolvers: Toward Fast Continuous-Depth Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add a learned residual correction to Euler | Neural ODE on MNIST · MNIST test | inference speed at comparable accuracy: 8 times faster | Helped |
| Use HyperEuler instead of Euler | Neural ODE · MNIST test | inference speed at comparable accuracy | Helped |
| Add a learned residual correction to Euler | Neural ODE · MNIST and CIFAR10 test | pareto efficiency at low NFEs | Helped |
| Use HyperEuler rather than midpoint or RK4 at low NFEs | Neural ODE · MNIST and CIFAR10 test | pareto efficiency | Helped |
| Use HyperEuler rather than higher-order methods at higher NFEs | Neural ODE · MNIST and CIFAR10 test | pareto efficiency | Mixed |
| Add a learned residual correction to Heun | FFJORD continuous normalizing flows · CNF sampling | sample quality at low NFEs: two NFEs | Helped |
| Use Heun instead of HyperHeun at the same NFE budget | FFJORD continuous normalizing flows · CNF sampling | sample quality | Hurt |
| Swap the midpoint base solver for other members of the second-order alpha family without finetuning | HyperMidpoint · MNIST test | terminal solution MAPE | Helped |
| Train HyperEuler with trajectory fitting | Galerkin Neural ODE · Periodic signal trajectory tracking | global truncation error | Helped |
| Use HyperEuler rather than midpoint or RK4 | Galerkin Neural ODE · Periodic signal trajectory tracking | pareto efficiency: 10 - 25 NFE | Helped |
Closest papers by meaning
- Implicit regularization of deep residual networks towards neural ODEs (ICLR 2024 · similarity 0.86)
- How Deep Do We Need: Accelerating Training and Inference of Neural ODEs via Control Perspective (ICML 2024 · similarity 0.86)
- NeuralSolver: Learning Algorithms For Consistent and Efficient Extrapolation Across General Tasks (NeurIPS 2024 · similarity 0.85)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 2024 · similarity 0.85)
- Improving Neural ODE Training with Temporal Adaptive Batch Normalization (NeurIPS 2024 · similarity 0.85)
- Layer Compression of Deep Networks with Straight Flows (AAAI 2024 · similarity 0.85)
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.