Research map / Architectures and efficiency / RNNs and normalization
Hidden Symmetries of ReLU Networks
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 |
|---|---|---|---|
| increase depth from 4 to 6 (fixed width) | randomly initialized ReLU networks · functional dimension experiments, n0=n1=...=n_{d-1} | fraction with maximal functional dimension (no hidden symmetries): 25%->1% at width 5; 66%->5% at width 15 | Hurt |
| increase width together with input dimension (depth fixed) | randomly initialized ReLU networks · functional dimension experiments, depth 4 and 6 | fraction with maximal functional dimension: depth 4: 25%->66% from width 5 to 15 | Helped |
| keep input dimension fixed at 5 and increase hidden width | randomly initialized ReLU networks · functional dimension experiments, n0=5 | fraction with maximal functional dimension: slight decrease | Hurt |
| theoretical: architectures with all hidden layers at least as wide as input admit a positive-measure set of parameters with no hidden symmetries | feedforward ReLU networks · theory (Theorem 1) | existence of parameters with no hidden symmetries | Helped |
Closest papers by meaning
- The Symmetries of Three-Layer ReLU Networks (NeurIPS 2026 · similarity 0.90)
- Most ReLU Networks Admit Identifiable Parameters (NeurIPS 2026 · similarity 0.89)
- Symmetry Induces Structure and Constraint of Learning (ICML 2024 · similarity 0.88)
- The Empirical Impact of Neural Parameter Symmetries, or Lack Thereof (NeurIPS 2024 · similarity 0.88)
- Parameter symmetries determine representational geometry in overparameterized nonlinear networks (NeurIPS 2026 · similarity 0.88)
- Symmetries in Overparametrized Neural Networks: A Mean Field View (NeurIPS 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.