Research map / Architectures and efficiency / RNNs and normalization
Initialization of ReLUs for Dynamical Isometry
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 |
|---|---|---|---|
| Replace He initialization with GSM parameter-sharing initialization | Fully connected ReLU networks · MNIST | classification accuracy | Helped |
| Replace He initialization with orthogonal submatrix parameter-sharing initialization | Fully connected ReLU networks · MNIST | classification accuracy | Helped |
| Replace He initialization with GSM parameter-sharing initialization | Fully connected ReLU networks · CIFAR-10 | classification accuracy | Helped |
| Replace He initialization with orthogonal submatrix parameter-sharing initialization | Fully connected ReLU networks · CIFAR-10 | classification accuracy | Helped |
| Use orthogonal rather than Gaussian submatrix weights | Parameter-sharing initialization · MNIST and CIFAR-10 | training reliability | Helped |
| Increase network depth under He initialization | Fully connected ReLU networks · MNIST and CIFAR-10 | classification accuracy | Hurt |
| Increase depth with Gaussian ReLU initialization | Deep ReLU Gaussian network ensembles · Signal cosine similarity | cosine similarity | Hurt |
Closest papers by meaning
- Sinusoidal Initialization, Time for a New Start (NeurIPS 2025 · similarity 0.89)
- Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer (ICML 2025 · similarity 0.86)
- From Lazy to Rich: Exact Learning Dynamics in Deep Linear Networks (ICLR 2025 · similarity 0.86)
- Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training (ICML 2025 · similarity 0.86)
- IDInit: A Universal and Stable Initialization Method for Neural Network Training (ICLR 2025 · similarity 0.86)
- Beyond IID weights: sparse and low-rank deep Neural Networks are also Gaussian Processes (ICLR 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.