Wei Lin

collaborators

2 papers

cs.LG2026

Hyperparameter Transfer Laws for Non-Recurrent Multi-Path Neural Networks

Shenxi Wu, Haosong Zhang, Xingjian Ma +4

Deeper modern architectures are costly to train, making hyperparameter transfer preferable to expensive repeated tuning. Maximal Update Parametrization ( $I^¼$ P) helps explain why ma…

cs.LG2025

Arithmetic-Mean $I^¼$ P for Modern Architectures: A Unified Learning-Rate Scale for CNNs and ResNets

Haosong Zhang, Shenxi Wu, Yichi Zhang +2

Choosing an appropriate learning rate remains a key challenge in scaling depth of modern deep networks. The classical maximal update parameterization ( $I^¼$ P) enforces a fixed per-l…

Hyperparameter Transfer Laws for Non-Recurrent Multi-Path Neural Networks

Arithmetic-Mean I^¼P for Modern Architectures: A Unified Learning-Rate Scale for CNNs and ResNets

Arithmetic-Mean $I^¼$ P for Modern Architectures: A Unified Learning-Rate Scale for CNNs and ResNets