Linear Recursive Feature Machines Dynamics Analyzed

Researchers have analyzed the dynamics and finite-sample trajectory recovery of linear Recursive Feature Machines (RFMs). In a noisy multi-output regression setting with isotropic sub-Gaussian input data and a low-rank teacher matrix, they extended the connection between linear RFMs and iteratively reweighted least squares. The study demonstrates that the learned feature matrix closely approximates its ideal counterpart, with an error decaying at a rate of O(sqrt(d/n)) with high probability for n samples. Experiments on real-world text and single-cell gene-expression data validated the features learned by this linear model. AI

IMPACT Provides theoretical insights into feature learning dynamics relevant for understanding and improving neural network training.

RANK_REASON Academic paper detailing theoretical analysis and experimental validation of a machine learning model. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• IArxiv
• linear RFMs
• Neural Networks
• Recursive Feature Machines
• ridge-regularized multi-output regression
• ScienceCast
• text mining

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →