Research map / Graph learning / Graph algorithms and clustering
Toward Deeper Understanding of Neural Networks: The Power of Initialization and a Dual View on Expressivity
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 ReLU activation instead of bounded activations in the random-network kernel approximation | random initialization of networks from computation skeletons · theoretical kernel approximation (ReLU skeleton) | required width r for kernel approximation | Helped |
| Stack many consecutive fully connected layers (depth increases) with a non-linear activation | fully connected skeleton kernel · theoretical limit of compositional kernel | limiting kernel form | Hurt |
| Replace two consecutive fully connected layers with a single layer with composed activation | fully connected skeletons · theoretical kernel equivalence | preserved dual kernel | No clear effect |
| Use random initialization with a small variance correction (bias-aware scaling) for the representation layer | random networks · theoretical kernel approximation | kernel approximation accuracy | No clear effect |
Closest papers by meaning
- An Empirical Study on The Properties of Random Bases for Kernel Methods (NeurIPS 2017 · similarity 0.87)
- From Lazy to Rich: Exact Learning Dynamics in Deep Linear Networks (ICLR 2025 · similarity 0.86)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.86)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.85)
- Position: A Theory of Deep Learning Must Include Compositional Sparsity (ICML 2025 · similarity 0.85)
- On the Impacts of the Random Initialization in the Neural Tangent Kernel Theory (NeurIPS 2024 · similarity 0.85)
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.