Research map / Label-efficient and robust learning / Recognition and long-tail
Decomposed Mutual Information Estimation for Contrastive Representation Learning
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 |
|---|---|---|---|
| Add conditional MI maximization to the InfoMin objective (ablation) | InfoMin (multi) · ImageNet-100 linear evaluation | accuracy: 77.2 -> 78.6 | Helped |
| Add conditional MI maximization to the InfoMin objective (ablation) | InfoMin (multi) · ImageNet-1K linear evaluation | accuracy: 70.2 -> 70.8 | Helped |
| Add conditional MI maximization | InfoMin · ImageNet-1K linear evaluation | accuracy: 70.1 -> 70.8 | Helped |
| Add conditional MI maximization | InfoMin · Stanford Cars transfer | accuracy: 48.1 -> 51.8 | Helped |
| Add conditional MI maximization | InfoMin · CUB transfer | accuracy: 41.7 -> 43.6 | Helped |
| Add conditional MI maximization | InfoMin · CIFAR-100 transfer | accuracy: 73.2 -> 75.0 | Helped |
| Add conditional MI maximization | InfoMin · Wizard of Wikipedia validation | perplexity: 18.85 -> 18.70 | Helped |
| Add conditional MI maximization | InfoMin · Wizard of Wikipedia validation | BLEU: 0.80 -> 0.82 | Helped |
| Replace standard crop subview with cutout (ablation) | DEMI · ImageNet-1K linear evaluation | accuracy: 70.5±0.1 | Hurt |
| Use two-way MI decompositions rather than only maximizing the decomposition with x′ as the subview (ablation) | DEMI · ImageNet linear evaluation | accuracy | Helped |
| Use the importance-sampling conditional critic approximation (ablation) | DEMI-IS · Synthetic Gaussian MI estimation | estimated mutual information | Hurt |
| Use the variational conditional MI approximation (ablation) | DEMI-VAR · Synthetic Gaussian MI estimation | estimated mutual information | Mixed |
Closest papers by meaning
- What Makes for Good Views for Contrastive Learning? (NeurIPS 2020 · similarity 0.89)
- Conditional Negative Sampling for Contrastive Learning of Visual Representations (ICLR 2021 · similarity 0.88)
- Multi-View Causal Representation Learning with Partial Observability (ICLR 2024 · similarity 0.86)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.86)
- Connecting Jensen–Shannon and Kullback–Leibler Divergences: A New Bound for Representation Learning (NeurIPS 2025 · similarity 0.85)
- Separating common from salient patterns with Contrastive Representation Learning (ICLR 2024 · similarity 0.85)
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.