Research map / Architectures and efficiency / Pruning and efficient tuning
A Bregman Learning Framework for Sparse Neural Networks
Leon Bungert, Tim Roith, Daniel Tenbrinck, Martin Burger
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 momentum (beta=0.9) to LinBreg (ablation) | LinBreg · Fashion-MNIST CNN, l1 | test accuracy: 89.2 -> 89.9 | Improved |
| use AdaBreg (Adam-style bias correction) instead of LinBreg (ablation) | LinBreg · Fashion-MNIST CNN, l1 | test accuracy: 89.2 -> 90.5 | Improved |
| use AdaBreg instead of LinBreg (ablation) | LinBreg · CIFAR-10 ResNet-18 | test accuracy: 90.9 -> 92.3 | Improved |
| apply post-training thresholding to LinBreg (ablation) | LinBreg · CIFAR-10 ResNet-18 | test accuracy and sparsity: 90.9 -> 90.2 | Mixed |
| apply post-training thresholding to AdaBreg (ablation) | AdaBreg · CIFAR-10 ResNet-18 | test accuracy and sparsity: 92.3 -> 90.5 | Mixed |
| add momentum (beta=0.9) to LinBreg (ablation) | LinBreg · CIFAR-10 ResNet-18 | test accuracy: 90.9 -> 90.4 | No clear effect |
| Bregman (AdaBreg) vs iterative pruning | iterative pruning · Fashion-MNIST CNN, l1 | N_total: 4.7 -> 2.3 | Improved |
| Bregman (AdaBreg) vs iterative pruning under group sparsity | pruning · Fashion-MNIST CNN, group sparsity | test accuracy: 86.9 -> 89.4 | Improved |
| LinBreg vs ProxGD Lasso | ProxGD Lasso · Fashion-MNIST CNN, l1 | N_total: 4.8 -> 1.9 | Improved |
| SGD + thresholding instead of ProxGD for Lasso | Lasso · Fashion-MNIST CNN, l1 | test accuracy: 89.4 -> 90.1 | Improved |
| use sparse initialization with 1% non-zero entries (r=0.01) | LinBreg · MNIST MLP | sparsity trajectory | Improved |
| increase l1 regularization parameter lambda | LinBreg · MNIST MLP | validation accuracy vs sparsity: 80% weights -> 10% weights | Mixed |
Closest papers by meaning
- A Bregman Proximal Viewpoint on Neural Operators (ICML 2025 · similarity 0.83)
- Sparse Cocktail: Every Sparse Pattern Every Sparse Ratio All At Once (ICML 2024 · similarity 0.82)
- Global Minimizers of $\ell^p$-Regularized Objectives Yield the Sparsest ReLU Neural Networks (NeurIPS 2025 · similarity 0.82)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.81)
- Two Sparse Matrices are Better than One: Sparsifying Neural Networks with Double Sparse Factorization (ICLR 2025 · similarity 0.81)
- Distributed Sparse Regression via Penalization (NeurIPS 2024 · similarity 0.81)
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.
