Research map / Probabilistic and scientific modeling / Variational and generative sampling
Improved Mutual Information Estimation
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 η-MINE instead of DV-MINE for MI-based pre-training | Deep InfoMax representation learning · STL10 test (pre-train STL10) | top-1 accuracy: 61.8 -> 63.3 | Helped |
| Use η-MINE instead of DV-MINE for MI-based pre-training | Deep InfoMax representation learning · CIFAR10 test (pre-train CIFAR10) | top-1 accuracy: 77.5 -> 74.8 | Hurt |
| Use η-MINE instead of DV-MINE for MI-based Jigsaw pre-training | ResNet50 · ImageNet 10% subset classification | top-1 accuracy: 8.5 ± 1.3 -> 11.0 ± 1.1 | Helped |
| Use η-MINE instead of DV-MINE for MI-based Jigsaw pre-training | ResNet50 · ImageNet 10% subset classification | top-5 accuracy: 20.0 ± 2.5 -> 24.1 ± 2.2 | Helped |
| Use η-MINE instead of DV-MINE for MI-based Jigsaw pre-training | ResNet50 · ImageNet 10% subset classification | top-10 accuracy: 27.8 ± 3.1 -> 32.7 ± 2.2 | Helped |
| Use η-MINE instead of DV-MINE for MI-regularized GAN training | Conditional GAN on 25-Gaussian data · 25-Gaussian mode recovery | generator steps to convergence: 140K -> 50K | Helped |
| Use η-MINE instead of existing MI-estimation baselines | Synthetic Gaussian MI estimation · 2-, 10- and 20-dimensional Gaussian data | MI estimation accuracy | Helped |
| Use the fixed-feature convex η-MINE formulation instead of neural η-MINE (ablation) | Synthetic Gaussian MI estimation · Gaussian data | MI estimation accuracy | Helped |
| Use η-MINE instead of DV-MINE for MI-based pre-training | Deep InfoMax representation learning · CIFAR10 test (pre-train STL10) | top-1 accuracy: 67.5 -> 68.3 | Helped |
| Use η-MINE instead of DV-MINE for MI-based pre-training | Deep InfoMax representation learning · STL10 test (pre-train CIFAR10) | top-1 accuracy: 55.1 -> 56.4 | Helped |
Closest papers by meaning
- Mutual Information Estimation via $f$-Divergence and Data Derangements (NeurIPS 2024 · similarity 0.91)
- Connecting Jensen–Shannon and Kullback–Leibler Divergences: A New Bound for Representation Learning (NeurIPS 2025 · similarity 0.91)
- Neural Mutual Information Estimation with Vector Copulas (NeurIPS 2025 · similarity 0.88)
- A Benchmark Suite for Evaluating Neural Mutual Information Estimators on Unstructured Datasets (NeurIPS 2024 · similarity 0.88)
- InfoNet: Neural Estimation of Mutual Information without Test-Time Optimization (ICML 2024 · similarity 0.86)
- MINDE: Mutual Information Neural Diffusion Estimation (ICLR 2024 · similarity 0.86)
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.