Research map / Label-efficient and robust learning / Noisy labels
Removing the Feature Correlation Effect of Multiplicative Noise
Zijun Zhang, Yining Zhang, Zongpeng Li
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 |
|---|---|---|---|
| Replace standard multiplicative noise with NCMN | CNN-16-10 · CIFAR-10 | test error (%): 3.76\pm 0.00 -> 3.41\pm 0.07 | Improved |
| Replace standard multiplicative noise with NCMN | CNN-16-10 · CIFAR-100 | test error (%): 18.08\pm 0.03 -> 17.55\pm 0.06 | Improved |
| Replace standard multiplicative noise with NCMN | WRN-22-7.5 · CIFAR-10 | test error (%): 3.59\pm 0.06 -> 3.02\pm 0.06 | Improved |
| Replace standard multiplicative noise with NCMN | WRN-22-7.5 · CIFAR-100 | test error (%): 18.60\pm 0.03 -> 16.70\pm 0.13 | Improved |
| Use NCMN-0, which injects multiplicative noise directly into normalized signal features (ablation) | NCMN · CIFAR-10, CNN-16-10 | test error (%): 3.41\pm 0.07 -> 3.51\pm 0.07 | Improved |
| Use NCMN-0 instead of NCMN-1 (ablation) | NCMN · CIFAR-100, CNN-16-10 | test error (%): 17.55\pm 0.06 -> 17.37\pm 0.05 | Improved |
| Apply NCMN noise once every two layers (NCMN-2) instead of every layer (ablation) | NCMN · CIFAR-10, CNN-16-10 | test error (%): 3.41\pm 0.07 -> 3.44\pm 0.03 | No clear effect |
| Apply NCMN noise once every two layers (NCMN-2) instead of every layer (ablation) | NCMN · CIFAR-100, CNN-16-10 | test error (%): 17.55\pm 0.06 -> 18.16\pm 0.04 | Hurt |
| Use NCMN-2 instead of NCMN-1 (ablation) | NCMN · CIFAR-10, WRN-22-7.5 | test error (%): 3.02\pm 0.06 -> 3.00\pm 0.05 | No clear effect |
| Use NCMN-2 instead of NCMN-1 (ablation) | NCMN · CIFAR-100, WRN-22-7.5 | test error (%): 17.09\pm 0.10 -> 16.70\pm 0.13 | Improved |
| Replace dropout with NCMN | LSTM language models · LSTM experiments | feature correlation | No clear effect |
Closest papers by meaning
- Dropout as a Structured Shrinkage Prior (ICML 2019 · similarity 0.87)
- Regularizing Deep Neural Networks by Noise: Its Interpretation and Optimization (NeurIPS 2017 · similarity 0.87)
- Information Dropout: Learning Optimal Representations Through Noisy Computation (TPAMI 2018 · similarity 0.86)
- Understanding the Disharmony Between Dropout and Batch Normalization by Variance Shift (CVPR 2019 · similarity 0.86)
- Improving robustness to corruptions with multiplicative weight perturbations (NeurIPS 2024 · similarity 0.85)
- Understanding and Mitigating the Label Noise in Pre-training on Downstream Tasks (ICLR 2024 · similarity 0.84)
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.
