No article text was recorded for this version.
maximal update parameterization
Version 2 of 2128 papers193 conceptsfrozen 2026-10-04
Papers in this version
- Blog: Scaling Diffusion Transformers Efficiently via $\mu$P
- Blog: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer
- Blog: $μ$pscaling small models: Principled warm starts and hyperparameter transfer
- Blog: Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit
- Blog: A Proof of Learning Rate Transfer under $\mu$P
- Blog: Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer
- Blog: u-$\mu$P: The Unit-Scaled Maximal Update Parametrization
- Blog: CompleteP for RL: Maintaining Feature Learning When Scaling Deep Reinforcement Learning
- Blog: Scaling Exponents Across Parameterizations and Optimizers
- Blog: μ-Parametrization for Mixture of Experts
- Blog: The Importance of Being Lazy: Scaling Limits of Continual Learning
- Blog: $\mu$PC: Scaling Predictive Coding to 100+ Layer Networks
- Blog: Function-Space Learning Rates
- Blog: Variational Deep Learning via Implicit Regularization
- Blog: A Stable Whitening Optimizer for Efficient Neural Network Training
- Blog: Tensor Programs IVb: Adaptive Optimization in the Infinite-Width Limit
- Blog: On the Residual Scaling of Looped Transformers: Stability and Transferability
- Blog: Extending $\mu$P: Spectral Conditions for Feature Learning Across Optimizers
- Blog: A thorough reproduction and evaluation of $\mu$P
- Blog: Maximal Update Parametrization and Zero-Shot Hyperparameter Transfer for Fourier Neural Operators
- Blog: Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer
- Blog: Tensor Programs IV: Feature Learning in Infinite-Width Neural Networks
- Blog: μ P2: Effective Sharpness Aware Minimization Requires Layerwise Perturbation Scaling
- Blog: Don't be lazy: CompleteP enables compute-efficient deep transformers
- Blog: Training Deep Learning Models with Norm-Constrained LMOs
- Blog: Infinite Limits of Multi-head Transformer Dynamics
- Blog: On Feature Learning in Structured State Space Models
- Blog: $\mu$pscaling small models: Principled warm starts and hyperparameter transfer
- Blog: Hyperparameter Transfer with Mixture-of-Expert Layers
- Blog: Training Deep Learning Models with Norm-Constrained LMOs
- Blog: $\boldsymbol{\mu}\mathbf{P^2}$: Effective Sharpness Aware Minimization Requires Layerwise Perturbation Scaling
- Blog: μPC: Scaling Predictive Coding to 100+ Layer Networks
- Blog: Training Deep Learning Models with Norm-Constrained LMOs
- Blog: Local Loss Optimization in the Infinite Width: Stable Parameterization of Predictive Coding Networks and Target Propagation
- Blog: Hyperparameter Transfer Enables Consistent Gains of Matrix-Preconditioned Optimizers Across Scales
- Blog: Scaling Diffusion Transformers Efficiently via μP
- Blog: Tensor Programs IV: Feature Learning in Infinite-Width Neural Networks
- Blog: μP2: Effective Sharpness Aware Minimization Requires Layerwise Perturbation Scaling
- Blog: Cerebras-GPT: Open Compute-Optimal Language Models Trained on the Cerebras Wafer-Scale Cluster
- Blog: On the Origin of Neural Scaling Laws: from Random Graphs to Natural Language
- Blog: On the Parameterization of Second-Order Optimization Effective towards the Infinite Width
- Blog: Communication-Efficient Language Model Training Scales Reliably and Robustly: Scaling Laws for DiLoCo
- Blog: HyperCLOVA X THINK Technical Report
- Blog: Non-Gaussian Tensor Programs
- Blog: Celo: Training Versatile Learned Optimizers on a Compute Diet
- Blog: Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit
- Blog: A Proof of Learning Rate Transfer under μP
- Blog: Mu-Parametrization for Mixture of Experts
- Blog: A Stable Whitening Optimizer for Efficient Neural Network Training
- Blog: Scalable Optimization in the Modular Norm
- Blog: Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer
- Blog: BTLM-3B-8K: 7B Parameter Performance in a 3B Parameter Model
- Blog: Research without Re-search: Maximal Update Parametrization Yields Accurate Loss Prediction across Scales
- Blog: Scaling Laws and Symmetry, Evidence from Neural Force Fields
- Blog: Hyperparameter Transfer Enables Consistent Gains of Matrix-Preconditioned Optimizers Across Scales
- Blog: Power Scheduler: A Batch Size and Token Number Agnostic Learning Rate Scheduler
- Blog: Adafactor: Adaptive Learning Rates with Sublinear Memory Cost
- Blog: FM4NPP: A Scaling Foundation Model for Nuclear and Particle Physics
- Blog: Optimal Scaling Needs Optimal Norm
- Blog: VaViM and VaVAM: Autonomous Driving through Video Generative Modeling
Showing 60 of 128 papers.
History & citation
Current · Permanent link
- width-dependent learning rate scaling split from this direction (56% overlap)
Cite this direction
Latest version (a living page):
@misc{lacuna_maximal_update_parameterization_99534441,
title = {maximal update parameterization},
author = {{Lacuna}},
howpublished = {Lacuna research direction},
year = {2026},
url = {https://lacuna.tiptreesystems.com/direction/maximal-update-parameterization/rdr_99534441e3874228b0c28ab4dcafda20},
note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}This exact version (version 2, frozen text):
@misc{lacuna_maximal_update_parameterization_99534441_v2,
title = {maximal update parameterization},
author = {{Lacuna}},
howpublished = {Lacuna research direction},
year = {2026},
url = {https://lacuna.tiptreesystems.com/direction/maximal-update-parameterization/rdr_99534441e3874228b0c28ab4dcafda20/v/2},
note = {Version 2, frozen 2026-10-04. Accessed 2026-11-22},
}