Research map / Architectures and efficiency / RNNs and normalization
On Enhancing Expressive Power via Compositions of Single Fixed-Size ReLU Network
Shijun Zhang, Jianfeng Lu, Hongkai Zhao
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 |
|---|---|---|---|
| increase composition count r of a fixed-size ReLU block (r=1 to r=4) (ablation) | L2∘g_n^{∘r}∘L1 RCNet, n=100 · Function approximation (target f on [0,1]^2) | MSE test loss: 1.22×10^-2 to 1.27×10^-5 (lower is better) | Improved |
| increase composition count r of a fixed-size ReLU block (r=1 to r=4) (ablation) | L2∘g_n^{∘r}∘L1 RCNet, n=200 · Function approximation (target f on [0,1]^2) | MSE test loss: 8.92×10^-3 to 1.11×10^-6 (lower is better) | Improved |
| increase composition count r of a fixed-size ReLU block (r=1 to r=4) (ablation) | L2∘g_n^{∘r}∘L1 RCNet, n=100 and n=200 · Function approximation (target f on [0,1]^2) | MAX test loss: 4.21×10^-1 to 2.77×10^-2 (n=100); 4.05×10^-1 to 8.61×10^-3 (n=200) | Improved |
| increase composition count r of a fixed-size ReLU block (r=1 to r=4) (ablation) | L2∘g_n^{∘r}∘L1 RCNet, n=30 · Spiral binary classification | test accuracy: 0.574781 to 0.874866 | Improved |
| increase composition count r of a fixed-size ReLU block (r=1 to r=4) (ablation) | L2∘g_n^{∘r}∘L1 RCNet, n=40 · Spiral binary classification | test accuracy: 0.575149 to 0.904041 | Improved |
Closest papers by meaning
- On the Expressiveness of Rational ReLU Neural Networks With Bounded Depth (ICLR 2025 · similarity 0.89)
- Characterizing ResNet's Universal Approximation Capability (ICML 2024 · similarity 0.89)
- Expressivity of ReLU-Networks under Convex Relaxations (ICLR 2024 · similarity 0.88)
- ReLU Network with Width $d+\mathcal{O}(1)$ Can Achieve Optimal Approximation Rate (ICML 2024 · similarity 0.88)
- Position: A Theory of Deep Learning Must Include Compositional Sparsity (ICML 2025 · similarity 0.87)
- Most ReLU Networks Admit Identifiable Parameters (NeurIPS 2026 · 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.
