Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Nested Dropout
Yufei Cui, Yu Peng Mao, Ziquan Liu, Qiao Li, Antoni B. Chan, Xue Liu, Tei‐Wei Kuo, Chun Jason Xue
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 |
|---|---|---|---|
| use nested BN3 instead of separately trained smallest-width SVB | BN3 · CIFAR-10 (VGG11) | ECE: 0.17 -> 0.10 | Improved |
| use full-width BN3 instead of MC inference over widths (ablation) | BN3 · CIFAR-10 (VGG11) | accuracy (%): 89.4 -> 92.6 | Improved |
| learn the nested dropout rate per node (BN3 vs FN3) (ablation) | BN3 · image classification | accuracy, ECE, AUPR | Improved |
| use BN3 instead of IBNN | MobileNetV2 · CIFAR/Tiny ImageNet | all metrics | Improved |
| use BN3 instead of IBNN | VGG11 · CIFAR/Tiny ImageNet | ECE | Hurt |
| add temperature scaling to FN3 | FN3 · image classification | calibration | Mixed |
| distil BN3 into SN3 with 25% channels and Bayesian last layer | VGG11 BN3 · Tiny ImageNet | accuracy (%): 54.00 -> 54.98 | Improved |
| distil BN3 into SN3 with 25% channels and Bayesian last layer | ResNeXt BN3 · Tiny ImageNet | accuracy (%): 61.19 -> 63.10 | Improved |
| train BN3 once instead of IBNN per width | BN3 · training time | hours: 14.9 h vs 7.5xN h | Improved |
| add VND ordering to VAE latent (100% dim) | VAE · CIFAR-10 generation | FID: 147.35 -> 114.12 | Improved |
| add VND ordering to VAE latent (100% dim) | VAE · 3D Chairs generation | reconstruction error: 1.380 -> 0.618 | Improved |
| add VND ordering to VAE latent (50% dim) | VAE · Chest X-ray generation | FID: 282.62 -> 285.79 | Hurt |
Closest papers by meaning
- Bayesian Nested Neural Networks for Uncertainty Calibration and Adaptive Compression (CVPR 2021 · similarity 0.90)
- Variational Bayesian Dropout With a Hierarchical Prior (CVPR 2019 · similarity 0.85)
- NestedNet: Learning Nested Sparse Structures in Deep Neural Networks (CVPR 2018 · similarity 0.84)
- Variational Dropout Sparsifies Deep Neural Networks (ICML 2017 · similarity 0.84)
- Nested Learning: The Illusion of Deep Learning Architectures (NeurIPS 2025 · similarity 0.83)
- Stochastic Modified Equations and Dynamics of Dropout Algorithm (ICLR 2024 · similarity 0.83)
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.
