Research map / Probabilistic and scientific modeling / Dynamical systems
Optimal Control Theoretic Neural Optimizer: From Backpropagation to Dynamic Programming
Guan-Horng Liu, Tianrong Chen, Evangelos Theodorou
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 |
|---|---|---|---|
| add layer-wise feedback policies (Qux term) to SGD/RMSprop/EKFAC-style updates (ablation) | OCNOpt variants · residual and inception networks | accuracy | Improved |
| add feedback policies under a large learning rate (ablation) | SGD baseline · DIGITS, MNIST, SVHN | convergence | Improved |
| use amortized curvature instead of vanilla DDP | DDP · FCN MNIST | accuracy: 90.42 -> 93.30 | Improved |
| use OCNOpt instead of E-MSA | E-MSA · CIFAR10 residual network | accuracy: 75.60 -> 85.85 | Improved |
| use OCNOpt instead of EKFAC | EKFAC · DIGITS FCN | accuracy: 95.24 -> 95.13 | No clear effect |
| use OCNOpt instead of EKFAC | EKFAC · MNIST residual network | accuracy: 98.77 -> 98.76 | No clear effect |
| use OCNOpt instead of SGD | SGD · SpoAD Neural ODE | accuracy: 97.70 -> 97.41 | Hurt |
| use OCNOpt instead of Adam | Adam · SVHN Neural ODE | accuracy: 91.92 -> 95.77 | Improved |
| use OCNOpt instead of Adam | Adam · Gas CNF | NLL: -6.42 -> -7.55 | Improved |
| add adaptive bandit alignment instead of fixed alignment (ablation) | OCNOpt · CIFAR10 (6000 iterations) | accuracy: 85.20 -> 85.65 | Improved |
| add random alignment instead of fixed alignment (ablation) | OCNOpt · CIFAR10 (6000 iterations) | accuracy: 85.20 -> 85.27 | Improved |
| add adaptive alignment on top of EKFAC baseline (ablation) | EKFAC · CIFAR10 (6000 iterations) | accuracy: 84.67 -> 85.65 | Improved |
Closest papers by meaning
- DDPNOpt: Differential Dynamic Programming Neural Optimizer (ICLR 2021 · similarity 0.93)
- Dynamic Game Theoretic Neural Optimizer (ICML 2021 · similarity 0.91)
- Neural Time-Reversed Generalized Riccati Equation (AAAI 2024 · similarity 0.88)
- Physics-Informed Neural Network Policy Iteration: Algorithms, Convergence, and Verification (ICML 2024 · similarity 0.87)
- Optimal Control Operator Perspective and a Neural Adaptive Spectral Method (AAAI 2025 · similarity 0.87)
- Deep learning for continuous-time stochastic control with jumps (NeurIPS 2025 · 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.
