Research map / Probabilistic and scientific modeling / Variational and generative sampling
Deep Variational Multivariate Information Bottleneck - A Framework for Variational Losses
Eslam Abdelaleem, Ilya Nemenman, K. Michael Martini
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 |
|---|---|---|---|
| add MINE-based mutual information decoder (DVSIB) instead of beta-VAE | beta-VAE · Noisy MNIST Y view (linear SVM) | accuracy %: 96.3 -> 97.8 | Improved |
| add beta trade-off weighting to DVCCA (ablation) | DVCCA · Noisy MNIST Y view (linear SVM) | accuracy %: 89.6 -> 95.4 | Improved |
| add beta weighting to DVCCA with joint input (ablation) | jDVCCA · Noisy MNIST joint [X,Y] (linear SVM) | accuracy %: 92.5 -> 96.7 | Improved |
| add private variables to beta-DVCCA (ablation) | beta-DVCCA · Noisy MNIST Y view (linear SVM) | accuracy %: 95.4 -> 95.5 | No clear effect |
| add private variables to DVSIB (ablation) | DVSIB · Noisy MNIST Y view (linear SVM) | accuracy %: 97.8 -> 97.8 | No clear effect |
| add private variables to DVCCA (ablation) | DVCCA · Noisy MNIST Y view (linear SVM) | accuracy %: 89.6 -> 92.1 | Improved |
| use DVIB (compress Y, predict X) instead of PCA | PCA · Noisy MNIST Y view (linear SVM) | accuracy %: 90.5 -> 90.4 | No clear effect |
| use DVSIB instead of beta-VAE as training set grows | beta-VAE · Noisy MNIST Y view, k_Z=64, sample-size scaling | log-log slope: 0.196 ± 0.013 -> 0.345 ± 0.007 | Improved |
| use joint [X,Y] input with beta-VAE-style DVCCA-p | beta-DVCCA-private · Noisy MNIST Y view with private variables | accuracy %: 95.5 -> 92.7 | Hurt |
Closest papers by meaning
- Differentiable Information Bottleneck for Deterministic Multi-view Clustering (CVPR 2024 · similarity 0.87)
- Information Bottleneck Analysis of Deep Neural Networks via Lossy Compression (ICLR 2024 · similarity 0.86)
- Cauchy-Schwarz Divergence Information Bottleneck for Regression (ICLR 2024 · similarity 0.85)
- Multilayer Matrix Factorization via Dimension-Reducing Diffusion Variational Inference (ICML 2025 · similarity 0.85)
- Learning Optimal Multimodal Information Bottleneck Representations (ICML 2025 · similarity 0.85)
- Information-theoretic Generalization Analysis for VQ-VAEs: A Role of Latent Variables (NeurIPS 2025 · 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.
