Symmetry-aware learning offers polynomial advantage in multi-index models

Researchers have established a polynomial sample complexity separation between symmetry-aware and symmetry-agnostic feature learning for multi-index models. Their analysis, focusing on high-dimensional Gaussian covariates with cyclic symmetry, reveals that architectural weight sharing and data augmentation exploiting symmetry achieve significantly better sample efficiency than learning without symmetry. This advantage is particularly pronounced for polynomial links with an information exponent of three or higher, where symmetry-aware methods require substantially fewer samples for directional recovery. AI

IMPACT This research could lead to more sample-efficient AI models by better leveraging inherent data symmetries.

RANK_REASON The item is an academic paper detailing theoretical research on feature learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• Charles Hermite
• cs.LG
• DagsHub
• Gaussian covariates
• Gotit.pub
• Hugging Face
• IArxiv
• multi-index models
• ScienceCast
• SGD
• Symmetry-Aware Feature Learning

• paper
• other

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