Research map / Architectures and efficiency / Pruning and efficient tuning
Coordinating Filters for Faster Deep Neural Networks
Wei Wen, Cong Xu, Chunpeng Wu, Yandan Wang, Yiran Chen, Hai 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 |
|---|---|---|---|
| add l2-norm Force Regularization during training | AlexNet PCA LRA · AlexNet (ImageNet) | average rank ratio (%): 72.29% -> 46.98% | Improved |
| add l1-norm Force Regularization during training | AlexNet PCA LRA · AlexNet (ImageNet) | average rank ratio (%): 72.29% -> 50.03% | Improved |
| add l2-norm Force Regularization during training | ConvNet · ConvNet (CIFAR-10) | average rank ratio (%): 74.48% -> 54.17% | Improved |
| add l2-norm Force Regularization before PCA decomposition | AlexNet lra · AlexNet conv3 (GPU) | speedup: 1.58x -> 2.16x | Improved |
| add l2-norm Force Regularization before PCA decomposition | AlexNet lra · AlexNet conv4 (CPU) | speedup: 1.60x -> 2.76x | Improved |
| add l2-norm Force Regularization on conv3-5 of lra2 | AlexNet lra2 · AlexNet ImageNet GPU conv3 | speedup: 1.89x -> 2.25x | Improved |
| use lra2 (3x3 shape preserved) instead of lra1 for conv3-5 | AlexNet · AlexNet ImageNet GPU conv3 | speedup: 0.86x -> 1.89x | Improved |
| scale regularization gradient by ||W_i|| instead of 1/||W_i|| (ablation) | Force Regularization · ConvNet (CIFAR-10) | conv3 rank (of 64): 32/64 -> 30/64 | Improved |
| use PCA or SVD instead of k-means for decomposition | AlexNet with l2-norm force · AlexNet (ImageNet) | top-1 error (%): 44.80% -> 43.25% | Improved |
| add Discrimination Regularization (reverse sign force) with l2-norm force | AlexNet · AlexNet (ImageNet) | top-1 error (%): 42.63% -> 41.71% | Improved |
| add Discrimination Regularization (reverse sign force) with l2-norm force | ResNets-20 · CIFAR-10 | top-1 error (%): 8.82% -> 7.97% | Improved |
| add Force Regularization before LRA when comparing at similar top-5 error to SSL | AlexNet lra2 · AlexNet ImageNet CPU | speedup: 3.13x -> 4.05x | Improved |
Closest papers by meaning
- Maestro: Uncovering Low-Rank Structures via Trainable Decomposition (ICML 2024 · similarity 0.86)
- COALA: Numerically Stable and Efficient Framework for Context-Aware Low-Rank Approximation (NeurIPS 2025 · similarity 0.86)
- Combining Explicit and Implicit Regularization for Efficient Learning in Deep Networks (NeurIPS 2022 · similarity 0.86)
- AdaRankGrad: Adaptive Gradient Rank and Moments for Memory-Efficient LLMs Training and Fine-Tuning (ICLR 2025 · similarity 0.85)
- Dynamical Low-Rank Compression of Neural Networks with Robustness under Adversarial Attacks (NeurIPS 2025 · similarity 0.85)
- Q3R: Quadratic Reweighted Rank Regularizer for Effective Low-Rank Training (NeurIPS 2025 · 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.
