Copertina

Gradient-Direction Sensitivity Reveals Linear-Centroid Coupling Hidden by Optimizer Trajectories

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autore principale:	Xu, Yongzhong
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Machine Learning Artificial Intelligence
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Spectral Edge Dynamics Reveal Functional Modes of Learning
di: Xu, Yongzhong
Pubblicazione: (2026)

Spectral Edge Dynamics of Training Trajectories: Signal--Noise Geometry Across Scales
di: Xu, Yongzhong
Pubblicazione: (2026)

When Do Attention Circuits Form? Developmental Trajectories of Capability and Attention-Sink Emergence Across Three 1B-ClassArchitectures
di: Xu, Yongzhong
Pubblicazione: (2026)

Low-Dimensional and Transversely Curved Optimization Dynamics in Grokking
di: Xu, Yongzhong
Pubblicazione: (2026)

Optimizer-Induced Low-Dimensional Drift and Transverse Dynamics in Transformer Training
di: Xu, Yongzhong
Pubblicazione: (2026)

Global Low-Rank, Local Full-Rank: The Holographic Encoding of Learned Algorithms
di: Xu, Yongzhong
Pubblicazione: (2026)

Early-Warning Signals of Grokking via Loss-Landscape Geometry
di: Xu, Yongzhong
Pubblicazione: (2026)

The Geometry of Multi-Task Grokking: Transverse Instability, Superposition, and Weight Decay Phase Structure
di: Xu, Yongzhong
Pubblicazione: (2026)

Spectral Probe-Circuits: A Three-Step Recipe for Identifying Attention-Head Circuits in Pretrained Transformers
di: Xu, Yongzhong
Pubblicazione: (2026)

Spectral Edge Dynamics: An Analytical-Empirical Study of Phase Transitions in Neural Network Training
di: Xu, Yongzhong
Pubblicazione: (2026)

Feature Repulsion and Spectral Lock-in: An Empirical Study of Two-Layer Network Grokking
di: Xu, Yongzhong
Pubblicazione: (2026)

Low-Dimensional Execution Manifolds in Transformer Learning Dynamics: Evidence from Modular Arithmetic Tasks
di: Xu, Yongzhong
Pubblicazione: (2026)

CSAttention: Centroid-Scoring Attention for Accelerating LLM Inference
di: Song, Chuxu, et al.
Pubblicazione: (2026)

Breaking the Reclustering Barrier in Centroid-based Deep Clustering
di: Miklautz, Lukas, et al.
Pubblicazione: (2024)

Explaining the Unexplained: Revealing Hidden Correlations for Better Interpretability
di: Jiang, Wen-Dong, et al.
Pubblicazione: (2024)

Learning with Noisy Labels through Learnable Weighting and Centroid Similarity
di: Wani, Farooq Ahmad, et al.
Pubblicazione: (2023)

Can Data-Driven Dynamics Reveal Hidden Physics? There Is A Need for Interpretable Neural Operators
di: Gao, Wenhan, et al.
Pubblicazione: (2025)

Robust Finite-Memory Policy Gradients for Hidden-Model POMDPs
di: Galesloot, Maris F. L., et al.
Pubblicazione: (2025)

Generalizing Behavior via Inverse Reinforcement Learning with Closed-Form Reward Centroids
di: Lazzati, Filippo, et al.
Pubblicazione: (2025)

Bregman Centroid Guided Cross-Entropy Method
di: Gu, Yuliang, et al.
Pubblicazione: (2025)

Memory Determines Learning Direction: A Theory of Gradient-Based Optimization in State Space Models
di: Guan, JingChuan, et al.
Pubblicazione: (2025)

Mitigating Bias in Facial Recognition Systems: Centroid Fairness Loss Optimization
di: Conti, Jean-Rémy, et al.
Pubblicazione: (2025)

When Fireflies Cluster; Enhancing Automatic Clustering via Centroid-Guided Firefly Optimization
di: Ariyaratne, MKA, et al.
Pubblicazione: (2026)

Entropy Centroids as Intrinsic Rewards for Test-Time Scaling
di: Zhao, Wenshuo, et al.
Pubblicazione: (2026)

Variance-Reduction Guidance: Sampling Trajectory Optimization for Diffusion Models
di: Xu, Shifeng, et al.
Pubblicazione: (2025)

On Revealing the Hidden Problem Structure in Real-World and Theoretical Problems Using Walsh Coefficient Influence
di: Przewozniczek, M. W., et al.
Pubblicazione: (2025)

Light-Weight Benchmarks Reveal the Hidden Hardware Cost of Zero-Shot Tabular Foundation Models
di: Gangwani, Ishaan, et al.
Pubblicazione: (2025)

AlphaGrad: Non-Linear Gradient Normalization Optimizer
di: Sane, Soham
Pubblicazione: (2025)

Worker Disagreement Reveals Sharp Directions in Local SGD
di: Dimlioglu, Tolga, et al.
Pubblicazione: (2026)

STFlow: Data-Coupled Flow Matching for Geometric Trajectory Simulation
di: Brinke, Kiet Bennema ten, et al.
Pubblicazione: (2025)

Proportional Fairness in Non-Centroid Clustering
di: Caragiannis, Ioannis, et al.
Pubblicazione: (2024)

Intelligence as Trajectory-Dominant Pareto Optimization
di: Khanh, Truong Xuan, et al.
Pubblicazione: (2026)

Prototype Training with Dual Pseudo-Inverse and Optimized Hidden Activations
di: Tucci, Mauro
Pubblicazione: (2025)

Counterfactual Evaluation Reveals Hidden Capability Profiles in Clinical LLMs and Agents
di: Turk, Matt
Pubblicazione: (2026)

Identifiability Matters: Revealing the Hidden Recoverable Condition in Unbiased Learning to Rank
di: Chen, Mouxiang, et al.
Pubblicazione: (2023)

Spatio-Temporal Trajectory Foundation Model - Recent Advances and Future Directions
di: Yang, Sean Bin, et al.
Pubblicazione: (2025)

Directly Aligning the Full Diffusion Trajectory with Fine-Grained Human Preference
di: Shen, Xiangwei, et al.
Pubblicazione: (2025)

Offline Trajectory Optimization for Offline Reinforcement Learning
di: Zhao, Ziqi, et al.
Pubblicazione: (2024)

PSMGD: Periodic Stochastic Multi-Gradient Descent for Fast Multi-Objective Optimization
di: Xu, Mingjing, et al.
Pubblicazione: (2024)

Gradient Extrapolation-Based Policy Optimization
di: Swapnil, Ismam Nur, et al.
Pubblicazione: (2026)