Research map / Learning theory and algorithms / Kernels and regression theory
Deep Limits and a Cut-Off Phenomenon for Neural Networks
Benny Avelin, Anders Karlsson
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 network width from 1 to 2 | random tanh network · TV distance to point mass at 0 | mixing time: 9 -> 11 | Hurt |
| replace TanH with SiLU activation | random network · TV distance to point mass at 0 | mixing time: 11 -> 7 | Mixed |
| use non-smooth ReLU activation | random network · cut-off phenomenon | presence of cut-off | Hurt |
| use heuristic 1/sqrt(N) uniform weight scaling | random tanh network · cut-off depth | layers to cut-off: 3 to 10 | No clear effect |
| normalize weights with sqrt(3)/sqrt(N) instead | random tanh network · cut-off depth | layers to cut-off | Mixed |
Closest papers by meaning
- Towards a General Theory of Infinite-Width Limits of Neural Classifiers (ICML 2020 · similarity 0.84)
- The Separation Capacity of Random Neural Networks (NeurIPS 2023 · similarity 0.84)
- Expected Gradients of Maxout Networks and Consequences to Parameter Initialization (ICML 2023 · similarity 0.84)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.83)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · similarity 0.82)
- Wide Neural Networks Trained with Weight Decay Provably Exhibit Neural Collapse (ICLR 2025 · similarity 0.82)
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.
