Research map / Architectures and efficiency / Pruning and efficient tuning
Denoised Variance-Based Pruning with Optimal Brain Bias Compensation
Geon Tack Lee, Choo Jaegul, Kang Eun Jeon
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 denoised variance scoring (Marchenko-Pastur) to VBP (ablation) | VBP · most DeiT/Swin/ConvNeXt models | Top-1 accuracy: gains of 1-2% | Improved |
| add denoised variance scoring (ablation) | VBP · DeiT-S and Swin-B | Top-1 accuracy: decrease of less than 1% | No clear effect |
| add OB2C weight update to VBP (ablation) | VBP · DeiT-T | Top-1 accuracy: nearly 14% absolute | Improved |
| add OB2C weight update to VBP (ablation) | VBP · remaining models | Top-1 accuracy: 4-7% | Improved |
| use DVBP + OB2C instead of VBP at higher pruning ratio | VBP · Swin-S | Top-1 accuracy gap: 2.28% at 40% -> 7.33% at 50% -> 25.49% at 60% | Improved |
| use DVBP + OB2C instead of VBP | VBP · ConvNeXt-T | Top-1 accuracy: +29.46% | Improved |
| use DVBP + OB2C instead of VBP | VBP · ConvNeXt-B | Top-1 accuracy: +14.64% | Improved |
| add denoising and OB2C pass over VBP | VBP · pruning time | time: approximately 2x VBP; 66.92 seconds on Swin-B | Hurt |
Closest papers by meaning
- Towards Meta-Pruning via Optimal Transport (ICLR 2024 · similarity 0.86)
- MDP: Multidimensional Vision Model Pruning with Latency Constraint (CVPR 2025 · similarity 0.85)
- ICP: Immediate Compensation Pruning for Mid-to-high Sparsity (CVPR 2025 · similarity 0.85)
- Efficient Noise Calculation in Deep Learning-based MRI Reconstructions (ICML 2025 · similarity 0.84)
- Optimal Brain Apoptosis (ICLR 2025 · similarity 0.84)
- Benchmarking Out-of-Distribution Generalization Capabilities of DNN-based Encoding Models for the Ventral Visual Cortex. (NeurIPS 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.
