Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors

• URL: http://arxiv.org/abs/2606.25971v1
• Date: Wed, 24 Jun 2026 15:40:26 GMT
• Title: Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors
• Abstract summary: We propose emphMagnitude-Direction-Direction (MD) Decoupling.<n>It factorizes each weight into a fixed-norm direction on a hypersphere and learnable per-row and per-column magnitude gains, updated at separate learning rates.<n>Across both Adam and Muon, MD Decoupling improves on well-tuned baselines, transfers the optimal LR across model width without retuning, and continues to help at scale on large Mixture-of-Experts models.
• Score: 110.10775872372047
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Modern neural network training relies on optimizers such as Adam and Muon which act on each weight matrix as a single object. Yet every weight matrix carries two distinct quantities -- a \emph{magnitude} and a \emph{direction} -- and all optimizers stepping in the matrix as a whole couple their dynamics: the directional change from an update depends on the current magnitude, while the magnitude drifts as a byproduct of learning the direction, so neither is governed directly by the learning rate. Typical training therefore leans on surrounding recipes such as weight decay and warmup to keep learning stable at scale, though these regulate the coupling only indirectly; other recent methods instead constrain the weight to a fixed-norm sphere, but add no learnable magnitude, leaving scale control to normalization layers alone. We propose \emph{Magnitude--Direction (MD) Decoupling}, an optimizer modification that factorizes each weight into a fixed-norm direction on a hypersphere and learnable per-row and per-column magnitude gains, updated at separate learning rates, all while the model still sees a single fused weight tensor. The method is agnostic to the base optimizer and removes the need for weight decay and warmup. Across both Adam and Muon, MD Decoupling improves on well-tuned baselines, transfers the optimal LR across model width without retuning, and continues to help at scale on large Mixture-of-Experts (MoE) models. Treating magnitude and direction as separately controlled quantities thus yields more predictable training dynamics and a simple, broadly applicable improvement to modern optimizers.

Related papers

• When Does Scale-Invariant Optimization Become Unstable? An Exact Schedule Law with Weight Decay [6.625270757109195]
  We show that a single quantity captures all schedule and decay forcing, while norm growth induces an opposing geometric self-quenching effect.<n>This yields a sharp boundary that cleanly separates contraction- and expansion-dominated effective learning rate regimes.
  arXiv  Detail & Related papers  (2026-09-08T17:46:17Z)
• Learning in Curved Weight Space:Exponential-Linear Weight Reparameterization for Improved Optimization [1.8613536568358355]
  textbfmethod combines a sign-aware symmetric-exponential pathway with an identity-like linear pathway.<n>We train transformers on OpenWebText over nine width$times$depth configurations.<n>shortmethod reaches matched validation loss in 1.32-1.49$times$ fewer training steps, with the largest widths seeing the biggest gains.
  arXiv  Detail & Related papers  (2026-07-10T20:48:10Z)
• Fantastic Pretraining Optimizers and Where to Find Them II: Hyperball Optimization [67.29294986811314]
  Matrix baseds such as Muon can substantially speed up language model pretraining, but their gains over AdamW are observed to shrink when using standard constant weight decay.<n>We propose Hyperball, a simple wrapper that addresses this issue.<n>On Qwen3 style models up to 1.2B parameters, Muon Hyperball achieves 20-30% token equivalent speedup over weight decay baselines.
  arXiv  Detail & Related papers  (2026-06-15T16:09:02Z)
• Weibull Weight-Scale Parameter Evolution under AdamW Training Dynamics [0.0]
  We study why the Weibull weight-scale parameter $(t)$ grows, overshoots, and then relaxes during AdamW training.<n>Near saturation, alignment and decay approach balance, explaining the transition from weight-scale growth to relaxation.
  arXiv  Detail & Related papers  (2026-06-11T03:50:14Z)
• SAW: Stage-Aware Dynamic Weighting for Multi-Objective Reinforcement Learning in Large Language Models [66.92079232778707]
  We propose Stage-Aware Dynamic Weighting (SAW) as a lightweight, algorithm-agnostic dynamic weighting mechanism.<n>SAW reweights each dimension's reward or advantage contribution by its relative informativeness within the batch.<n> Experiments on tool-calling and text summarization tasks demonstrate that SAW consistently improves both training efficiency and final performance.
  arXiv  Detail & Related papers  (2026-06-05T10:00:19Z)
• Neural Network Optimization Reimagined: Decoupled Techniques for Scratch and Fine-Tuning [49.751529745537546]
  We propose DualOpt, a novel approach that decouples optimization techniques specifically tailored for training from scratch.<n>For training from scratch, we introduce real-time layer-wise weight decay, designed to enhance both convergence and generalization.<n>We extend the layer-wise weight decay to dynamically adjust the rollback levels across layers, adapting to the varying demands of different downstream tasks.
  arXiv  Detail & Related papers  (2026-04-21T06:27:18Z)
• Controlled LLM Training on Spectral Sphere [76.60985966206746]
  We introduce the textbfSpectral Sphere algorithm (SSO), which enforces strict module-wise spectral constraints on both weights and their updates.<n>We observe significant practical stability benefits, including improved MoE router load balancing, suppressed outliers, and strictly bounded activations.
  arXiv  Detail & Related papers  (2026-01-13T09:59:47Z)
• Revisiting Gradient Descent: A Dual-Weight Method for Improved Learning [4.751362812627724]
  We introduce a novel framework for learning in neural networks by decomposing each neuron's weight vector into two distinct parts.<n>We show that this decomposition enhances generalization and resists overfitting, especially when training data are sparse or noisy.
  arXiv  Detail & Related papers  (2025-03-15T02:32:47Z)
• Rotational Equilibrium: How Weight Decay Balances Learning Across Neural Networks [33.88586668321127]
  This study investigates how weight decay affects the update behavior of individual neurons in deep neural networks. We show that explicitly controlling the rotation provides the benefits of weight decay while substantially reducing the need for learning rate warmup.
  arXiv  Detail & Related papers  (2023-05-26T19:14:01Z)
• AdamP: Slowing Down the Slowdown for Momentum Optimizers on Scale-invariant Weights [53.8489656709356]
  Normalization techniques are a boon for modern deep learning. It is often overlooked, however, that the additional introduction of momentum results in a rapid reduction in effective step sizes for scale-invariant weights. In this paper, we verify that the widely-adopted combination of the two ingredients lead to the premature decay of effective step sizes and sub-optimal model performances.
  arXiv  Detail & Related papers  (2020-06-15T08:35:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.