Delayed Optimizer-State Transport Shapes Short-Horizon Training Decisions
cs.LG, physics.comp-ph
Submitted: 2026-08-25
Updated: 2026-08-25
Comments: 24 pages, 14 figures, 3 tables
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Adam: A Method for Stochastic Optimization
- Decoupled Weight Decay Regularization
- A Survey on Data Selection for Language Models
- Rethinking Data Mixture for Large Language Models: A Comprehensive Survey and New Perspectives
- DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining
- RegMix: Data Mixture as Regression for Language Model Pre-training
- Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models
- Efficient Online Data Mixing For Language Model Pre-Training
- Dynamic Gradient Alignment for Online Data Mixing
- PiKE: Adaptive Data Mixing for Large-Scale Multi-Task Learning Under Low Gradient Conflicts
- Estimating Training Data Influence by Tracing Gradient Descent
- HYDRA: Hypergradient Data Relevance Analysis for Interpreting Deep Neural Networks
- Commute Your Domains: Trajectory Optimality Criterion for Multi-Domain Learning
- Optimizer Memory Makes Shuffle Order a First-Order Source of Fine-Tuning Noise
- In-Run Data Shapley for Adam Optimizer
- Analysing Mathematical Reasoning Abilities of Neural Models
- Group Normalization
- Anisotropy and universality: Critical Binder cumulant of the two-dimensional Ising model
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks