Research map / Optimization / Optimizers
A Trace-restricted Kronecker-Factored Approximation to Natural Gradient
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 |
|---|---|---|---|
| Use trace-restricted Kronecker factors with normal damping instead of KFAC | VGG16 · CIFAR-10 | testing accuracy: 92.70$\pm$ 0.39 -> 93.33$\pm$ 0.20 | Helped |
| Use trace-restricted Kronecker factors with normal damping instead of KFAC | VGG16 · CIFAR-100 | testing accuracy: 70.84$\pm$ 0.27 -> 71.06$\pm$0.45 | Helped |
| Use trace-restricted Kronecker factors with normal damping instead of KFAC | ResNet20 · CIFAR-10 | testing accuracy: 93.69$\pm$ 0.16 -> 93.76$\pm$ 0.19 | No clear effect |
| Use trace-restricted Kronecker factors with normal damping instead of KFAC | ResNet20 · CIFAR-100 | testing accuracy: 74.18$\pm$ 0.27 -> 74.43$\pm$ 0.41 | No clear effect |
| Replace normal damping with the new trace-restricted damping (ablation) | TKFAC on VGG16 · CIFAR-10 | testing accuracy: 93.33$\pm$ 0.20 -> 93.42$\pm$0.11 | No clear effect |
| Replace normal damping with the new trace-restricted damping (ablation) | TKFAC on VGG16 · CIFAR-100 | testing accuracy: 71.06$\pm$0.45 -> 70.91$\pm$ 0.34 | No clear effect |
| Replace normal damping with the new trace-restricted damping (ablation) | TKFAC on ResNet20 · CIFAR-10 | testing accuracy: 93.76$\pm$ 0.19 -> 94.66$\pm$ 0.12 | Helped |
| Replace normal damping with the new trace-restricted damping (ablation) | TKFAC on ResNet20 · CIFAR-100 | testing accuracy: 74.43$\pm$ 0.41 -> 77.01$\pm$ 0.34 | Helped |
| Use the new trace-restricted damping instead of KFAC | ResNet20 · CIFAR-10 | testing accuracy: 93.69$\pm$ 0.16 -> 94.66$\pm$ 0.12 | Helped |
| Use the new trace-restricted damping instead of KFAC | ResNet20 · CIFAR-100 | testing accuracy: 74.18$\pm$ 0.27 -> 77.01$\pm$ 0.34 | Helped |
Closest papers by meaning
- A Kronecker-factored approximate Fisher matrix for convolution layers (ICML 2016 · similarity 0.92)
- Kronecker-factored Curvature Approximations for Recurrent Neural Networks (ICLR 2018 · similarity 0.92)
- Kronecker-Factored Approximate Curvature for Physics-Informed Neural Networks (NeurIPS 2024 · similarity 0.92)
- Provable Acceleration of Nesterov's Accelerated Gradient for Asymmetric Matrix Factorization and Linear Neural Networks (NeurIPS 2024 · similarity 0.86)
- Global curvature for second-order optimization of neural networks (ICML 2025 · similarity 0.85)
- AdaFisher: Adaptive Second Order Optimization via Fisher Information (ICLR 2025 · similarity 0.84)
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.