Research map / Architectures and efficiency / RNNs and normalization
Regularizing by the Variance of the Activations' Sample-Variances
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 VCL to ELU | ELU network · CIFAR-10 | test error: 0.0698 -> 0.0615 | Helped |
| Add VCL to ELU | ELU network · CIFAR-100 | test error: 0.287 -> 0.256 | Helped |
| Add batchnorm to ReLU | ReLU network · CIFAR-10 | test error: 0.0836 -> 0.0778 | Helped |
| Add VCL to ReLU | ReLU network · CIFAR-100 | test error: 0.328 -> 0.303 | Helped |
| Add VCL to Leaky ReLU | Leaky ReLU network · CIFAR-10 | test error: 0.0670 -> 0.0645 | Helped |
| Add batchnorm to Leaky ReLU | Leaky ReLU network · CIFAR-10 | test error: 0.0670 -> 0.0708 | Hurt |
| Add VCL to the Deep ELU network | Deep ELU network · Tiny ImageNet validation | validation error: 0.392 -> 0.373 | Helped |
| Add batchnorm to the Deep ELU network | Deep ELU network · Tiny ImageNet validation | validation error: 0.392 -> 0.402 | Hurt |
| Add VCL | UCI networks with ReLU · UCI datasets | number of wins: 15 -> 27 | Helped |
| Add VCL | UCI networks with ELU · UCI datasets | number of wins: 16 -> 23 | Helped |
| Add VCL | UCI networks with SELU · UCI datasets | number of wins: 15 -> 28 | Helped |
| Reduce VCL sample size from n=9 to n=2 (ablation) | ELU network · CIFAR-100 | test error: 0.276 -> 0.256 | Helped |
Closest papers by meaning
- Deconstructing the Regularization of BatchNorm (ICLR 2021 · similarity 0.90)
- Understanding Batch Normalization (NeurIPS 2018 · similarity 0.90)
- Beyond BatchNorm: Towards a Unified Understanding of Normalization in Deep Learning (NeurIPS 2021 · similarity 0.90)
- Towards Understanding Regularization in Batch Normalization (ICLR 2019 · similarity 0.90)
- Regularizing activations in neural networks via distribution matching with the Wasserstein metric (ICLR 2020 · similarity 0.89)
- Towards Training Without Depth Limits: Batch Normalization Without Gradient Explosion (ICLR 2024 · similarity 0.89)
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.