Research map / Learning theory and algorithms / Kernels and regression theory
Deterministic Bounds and Random Estimates of Metric Tensors on Neuromanifolds
Ke Sun
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 Hutchinson unbiased estimate instead of empirical FIM | diagonal FIM estimation · MNLI (RoBERTa-base) | RelMAE: 53.9 -> 0.16 | Improved |
| use Hutchinson unbiased estimate instead of empirical FIM | diagonal FIM estimation · SpeechCommands (Wav2Vec2-base) | RelMAE: 56.8 -> 0.17 | Improved |
| use Hutchinson unbiased estimate instead of empirical FIM | diagonal FIM estimation · SST-2 (DistilBERT) | RelMAE: 1.15 -> 0.18 | Improved |
| use Hutchinson unbiased estimate instead of empirical FIM | diagonal FIM estimation · CIFAR-100 (ResNet-50) | RelMAE: 0.17 -> 0.11 | Improved |
| use low-rank estimate LR instead of Hutchinson | diagonal FIM estimation · SST-2 and MNLI (fine-tuned) | RelMAE: 0.18 -> 0.05 (SST-2); 0.16 -> 0.11 (MNLI) | Improved |
| use low-rank estimate LR instead of Hutchinson | diagonal FIM estimation · CIFAR-100 (ResNet-50) | RelMAE: 0.11 -> 0.97 | Hurt |
| use Hutchinson estimate instead of empirical FIM for speed | diagonal FIM estimation · SST-2 (DistilBERT) | speedup: x1 -> x1.07 | No clear effect |
| use low-rank LR(2) instead of empirical FIM for speed | diagonal FIM estimation · CIFAR-100 (ResNet-50) | speedup: x1 -> x0.46 | Hurt |
Closest papers by meaning
- Riemannian Metric Learning via Optimal Transport (ICLR 2023 · similarity 0.86)
- Riemannian Metric Matching for Scalable Geometric Modeling of Distributions (ICML 2026 · similarity 0.86)
- Learning on a Razor’s Edge: Identifiability and Singularity of Polynomial Neural Networks (ICLR 2026 · similarity 0.85)
- Geometry-Aware Similarity Metrics for Neural Representations on Riemannian and Statistical Manifolds (NeurIPS 2026 · similarity 0.84)
- RMLR: Extending Multinomial Logistic Regression into General Geometries (NeurIPS 2024 · similarity 0.84)
- Estimating Shape Distances on Neural Representations with Limited Samples (ICLR 2024 · 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.
