AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Scaling depth capacity via zero/one-layer model expansion

Zhiqi Bu

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Neural scaling laws · 0 citations (OpenAlex)

Read the paper

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 changedWhereMeasuredEffect
initialize new layers randomly or by copying instead of zeros (ablation)progressive depth expansion · GPT2/LLAMA3/Qwen3/DeepSeekV3/Mixtralvalidation loss / compute efficiency: not givenImproved
zero-initialize new layers (ablation)progressive depth expansion · GPT2 1-layer to 12-layervalidation loss / compute efficiency: not givenHurt
copy all layers (stack or interpolate) instead of copying only the last layer (ablation)multi-layer expansion · GPT2 6-layer, ResNet50validation loss / compute efficiency: not givenImproved
copying interpolate vs copying stack (ablation)multi-layer expansion · GPT2 6-layervalidation loss / compute efficiency: almost indistinguishableNo clear effect
use WSD schedule instead of cosine schedule (ablation)zero-layer progressive training · GPT2 and ResNetvalidation loss / compute efficiency: small model catches up until tau 0.8T vs 0.5T (GPT)Improved
expand depth later in the stable phase of WSD (ablation)progressive training · GPT2 1-layervalidation loss / compute efficiency: mixing time about 30k iterations at 10% and 60% horizonNo clear effect
expand at 80% of iterations after decay begins (ablation)one-layer progressive training · GPT2 12-layervalidation loss / compute efficiency: cannot mix wellHurt
use multi-stage expansion (0 to 2 to 12) instead of single-stage (ablation)progressive training · GPT2 12-layervalidation loss / compute efficiency: similar final lossNo clear effect
expand from more than 2 layers instead of zero/one layer (ablation)progressive training · GPT2 12/24/36-layervalidation loss / compute efficiency: not givenImproved
use muP-scaled learning rate with Muon-NSGD (ablation)progressive training · GPT2 and ResNetvalidation loss / compute efficiency: optimal learning rate transfers across sizesImproved
apply zero/one-layer progressive training to MoEDeepSeekV3 and Mixtral · OpenWebTextvalidation loss / compute efficiency: 3.4x to 5.1xImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy