Research map / Architectures and efficiency / RNNs and normalization
Globally Gated Deep Linear Networks
Qianyi Li, Haim Sompolinsky
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 |
|---|---|---|---|
| Use finite-width kernel shape renormalization instead of the GP kernel | GGDLN · MNIST | generalization error | Improved |
| Use pretrained unsupervised soft k-means gatings instead of random gatings | GGDLN · MNIST | error rate | Improved |
| Increase the number of gatings | GGDLN · MNIST | error rate | Improved |
| Increase network width | GGDLN with localized receptive fields · ReLU teacher task | generalization error | Improved |
| Increase network width | GGDLN with localized receptive fields · ReLU teacher task with preferred inputs | generalization error | Hurt |
| Increase regularization amplitude sigma in the regime where bias falls and variance rises | GGDLN · MNIST | error rate | Mixed |
| Increase regularization amplitude sigma in the regime where bias and variance both rise | GGDLN · MNIST | error rate | Hurt |
| Use finite-width kernel renormalization | Deep GGDLN · Deep GGDLN task | bias | Improved |
| Increase network width | GGDLN with bottom-up gatings · Permuted MNIST | error rate | Improved |
| Increase gating threshold | GGDLN with bottom-up gatings · Permuted MNIST | error rate | Mixed |
| Increase network width | GGDLN with combined top-down and bottom-up gatings · Permuted and un-permuted MNIST tasks | generalization error | Hurt |
| Increase network depth | GGDLN with renormalized kernel · Deep GGDLN task | kernel flattening | Improved |
Closest papers by meaning
- Gaussian Gated Linear Networks (NeurIPS 2020 · similarity 0.89)
- Generalization of Scaled Deep ResNets in the Mean-Field Regime (ICLR 2024 · similarity 0.86)
- Theoretical Characterisation of the Gauss Newton Conditioning in Neural Networks (NeurIPS 2024 · similarity 0.86)
- Exact, Tractable Gauss-Newton Optimization in Deep Reversible Architectures Reveal Poor Generalization (NeurIPS 2024 · similarity 0.86)
- Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer (ICML 2025 · similarity 0.86)
- Global Convergence and Rich Feature Learning in $L$-Layer Infinite-Width Neural Networks under $\mu$ Parametrization (ICML 2025 · 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.
