Research map / Probabilistic and scientific modeling / Variational and generative sampling
Understanding the Limitations of Variational Mutual Information Estimators
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 |
|---|---|---|---|
| clip density ratios with SMILE (τ = 1.0) instead of using NWJ | discriminative MI estimation · Gaussian (MI 10) | MSE: 28.43 -> 0.31 | Helped |
| clip density ratios with SMILE (τ = 1.0) instead of using NWJ | discriminative MI estimation · Cubic (MI 10) | MSE: 27.00 -> 4.98 | Helped |
| clip density ratios with SMILE (τ = 1.0) instead of using NWJ | discriminative MI estimation · Gaussian (MI 10) | variance: 99.0 -> 0.34 | Helped |
| clip density ratios with SMILE (τ = 1.0) instead of using NWJ | discriminative MI estimation · Cubic (MI 10) | variance: 3.23 -> 0.30 | Helped |
| increase τ from 1.0 to 5.0 (ablation) | SMILE · Gaussian (MI 10) | MSE: 0.31 -> 1.26 | Hurt |
| increase τ from 1.0 to 10.0 (ablation) | SMILE · Gaussian (MI 10) | MSE: 0.31 -> 1.33 | Hurt |
| remove finite clipping by setting τ = ∞ (ablation) | SMILE · Gaussian (MI 10) | MSE: 0.31 -> 1.62 | Hurt |
| increase τ from 1.0 to 5.0 (ablation) | SMILE · Cubic (MI 10) | MSE: 4.98 -> 4.92 | No clear effect |
| increase τ from 1.0 to 10.0 (ablation) | SMILE · Cubic (MI 10) | MSE: 4.98 -> 9.91 | Hurt |
| remove finite clipping by setting τ = ∞ (ablation) | SMILE · Cubic (MI 10) | MSE: 4.98 -> 8.81 | Hurt |
| use a generative flow estimator instead of SMILE | MI estimation · Gaussian (MI 10) | MSE: 0.31 -> 0.22 | Helped |
| use a generative flow estimator instead of SMILE | MI estimation · Cubic (MI 10) | MSE: 4.98 -> 13.67 | Hurt |
Closest papers by meaning
- Mutual Information Estimation via $f$-Divergence and Data Derangements (NeurIPS 2024 · similarity 0.92)
- A Benchmark Suite for Evaluating Neural Mutual Information Estimators on Unstructured Datasets (NeurIPS 2024 · similarity 0.89)
- Neural Mutual Information Estimation with Vector Copulas (NeurIPS 2025 · similarity 0.88)
- InfoNet: Neural Estimation of Mutual Information without Test-Time Optimization (ICML 2024 · similarity 0.88)
- Mutual Information Estimation via Normalizing Flows (NeurIPS 2024 · similarity 0.88)
- MINDE: Mutual Information Neural Diffusion Estimation (ICLR 2024 · similarity 0.88)
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.