Research map / Architectures and efficiency / Pruning and efficient tuning
DeepHoyer: Learning Sparser Neural Network with Differentiable Scale-Invariant Sparsity Measures
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 |
|---|---|---|---|
| Use Hoyer-Square instead of Hoyer regularization (ablation) | LeNet-300-100 · MNIST | nonzero weights: 6.0k -> 4.6k | Helped |
| Use Hoyer-Square instead of Hoyer regularization (ablation) | LeNet-5 · MNIST | nonzero weights: 4.0k -> 3.5k | Helped |
| Apply Hoyer-Square regularization during training | AlexNet · ImageNet | parameters: 2.9M -> 2.85M | Helped |
| Apply Group-HS structural regularization | LeNet-300-100 · MNIST | FLOPs: 20.5k (7.70%) -> 16.5k (6.19%) | Helped |
| Apply Group-HS structural regularization | LeNet-5 · MNIST | FLOPs: 153.4k (6.69%) -> 169.9k (7.41%) | Mixed |
| Increase Group-HS regularization strength | ResNet-50 · ImageNet | FLOPs reduction: $1.89 \times$ -> $5.48 \times$ | Mixed |
| Increase Group-HS regularization strength | ResNet-56 · CIFAR-10 | FLOPs reduction: $2.38 \times$ -> $5.63 \times$ | Mixed |
| Increase Group-HS regularization strength | ResNet-110 · CIFAR-10 | FLOPs reduction: $2.30 \times$ -> $5.84 \times$ | Mixed |
Closest papers by meaning
- Two Sparse Matrices are Better than One: Sparsifying Neural Networks with Double Sparse Factorization (ICLR 2025 · similarity 0.88)
- Deep Weight Factorization: Sparse Learning Through the Lens of Artificial Symmetries (ICLR 2025 · similarity 0.88)
- Global Minimizers of $\ell^p$-Regularized Objectives Yield the Sparsest ReLU Neural Networks (NeurIPS 2025 · similarity 0.88)
- A Single-Step, Sharpness-Aware Minimization is All You Need to Achieve Efficient and Accurate Sparse Training (NeurIPS 2024 · similarity 0.87)
- SequentialAttention++ for Block Sparsification: Differentiable Pruning Meets Combinatorial Optimization (NeurIPS 2024 · similarity 0.87)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.87)
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.