Research map / Optimization / Optimizers
Not All Samples Are Created Equal: Deep Learning with Importance Sampling
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 upper-bound gradient-norm importance sampling instead of uniform sampling | Wide ResNet 28-2 · CIFAR10 | test error: 0.087 -> 0.079 | Helped |
| Use upper-bound gradient-norm importance sampling instead of uniform sampling | Wide ResNet 28-2 · CIFAR100 | test error: 0.34 -> 0.32 | Helped |
| Use upper-bound gradient-norm importance sampling instead of uniform sampling | ResNet-50 fine-tuning · MIT67 | test error: 33.74% -> 28.06% | Helped |
| Use upper-bound gradient-norm importance sampling instead of uniform sampling | LSTM · Pixel-by-pixel MNIST | test error: 0.1139 -> 0.1055 | Helped |
| Use upper-bound gradient-norm scores instead of loss scores for importance sampling (ablation) | Wide ResNet · CIFAR100 | sum of squared error: 0.017 -> 0.002 | Helped |
| Sample examples proportionally to loss instead of uniformly (ablation) | LSTM · Pixel-by-pixel MNIST | convergence | Hurt |
| Set presampling size B to 384 (ablation) | Upper-bound importance sampling · CIFAR10 | training loss: 0.6 | Helped |
| Increase presampling size (ablation) | Upper-bound importance sampling · CIFAR10 | training loss | Helped |
| Use loss-based rather than upper-bound importance scores (ablation) | Deep neural network training · Early training | gradient variance | Hurt |
Closest papers by meaning
- Optimizing importance weighting in the presence of sub-population shifts (ICLR 2025 · similarity 0.86)
- Efficient Backpropagation with Variance Controlled Adaptive Sampling (ICLR 2024 · similarity 0.85)
- Accelerating Neural Field Training via Soft Mining (CVPR 2024 · similarity 0.85)
- LEAPS: A discrete neural sampler via locally equivariant networks (ICML 2025 · similarity 0.85)
- Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty (ICML 2025 · similarity 0.84)
- Reparameterized Importance Sampling for Robust Variational Bayesian Neural Networks (ICML 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.