Research map / Architectures and efficiency / RNNs and normalization
From Hard to Soft: Understanding Deep Network Nonlinearities via Vector Quantization and Statistical Inference
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 |
|---|---|---|---|
| Add an orthogonality penalty to the filters (ablation) | largeCNN · SVHN (LR=0.001) | accuracy: 94.3 ± 0.1 -> 94.6 ± 0.2 | Helped |
| Add an orthogonality penalty to the filters (ablation) | largeCNN · SVHN (LR=0.0005) | accuracy: 94.4 ± 0.1 -> 95.0 ± 0.2 | Helped |
| Add an orthogonality penalty to the filters (ablation) | largeCNN · SVHN (LR=0.0001) | accuracy: 93.4 ± 0.0 -> 93.8 ± 0.1 | Helped |
| Add an orthogonality penalty to the filters (ablation) | largeCNN · CIFAR10 (LR=0.001) | accuracy: 80.3 ± 0.4 -> 84.0 ± 0.3 | Helped |
| Add an orthogonality penalty to the filters (ablation) | largeCNN · CIFAR10 (LR=0.0005) | accuracy: 80.2 ± 0.2 -> 82.3 ± 0.1 | Helped |
| Add an orthogonality penalty to the filters (ablation) | largeCNN · CIFAR10 (LR=0.0001) | accuracy: 76.2 ± 0.3 -> 79.1 ± 0.2 | Helped |
| Add an orthogonality penalty to the filters (ablation) | largeCNN · CIFAR100 (LR=0.001) | accuracy: 43.6 ± 0.2 -> 46.1 ± 0.2 | Helped |
| Add an orthogonality penalty to the filters (ablation) | largeCNN · CIFAR100 (LR=0.0005) | accuracy: 44.1 ± 0.4 -> 46.3 ± 0.2 | Helped |
| Add an orthogonality penalty to the filters (ablation) | largeCNN · CIFAR100 (LR=0.0001) | accuracy: 37.5 ± 0.5 -> 42.1 ± 0.3 | Helped |
| Use exact Gram-Schmidt orthogonalization | largeCNN with swish activation · CIFAR100 | accuracy: 61.2% | Helped |
Closest papers by meaning
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.87)
- From Alexnet to Transformers: Measuring the Non-linearity of Deep Neural Networks with Affine Optimal Transport (CVPR 2025 · similarity 0.87)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · similarity 0.87)
- MaxQ: Multi-Axis Query for N:M Sparsity Network (CVPR 2024 · similarity 0.86)
- Smooth Min-Max Monotonic Networks (ICML 2024 · similarity 0.86)
- Deep linear networks for regression are implicitly regularized towards flat minima (NeurIPS 2024 · similarity 0.86)
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.