New framework enables machine learning on variable-sized inputs

Researchers have introduced a novel framework for machine learning models that can handle inputs of varying sizes, such as point clouds or sequences of different lengths. This approach utilizes random sampling maps, generalizing methods like sampling with replacement and random binning, to compare and approximate inputs of different sizes. The framework provides explicit rates for generalization and sketching, applicable to functions defined on sequences, graphs, and tensors, including moment polynomials, homomorphism densities, permutation-invariant transformers, and graph neural networks. AI

IMPACT Enables more flexible and scalable machine learning models for diverse data types.

RANK_REASON Academic paper introducing a new theoretical framework for machine learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Any-Dimensional Learning by Sampling
• arXiv
• graph neural networks
• homomorphism densities
• machine learning
• moment polynomials
• permutation-invariant transformers
• phylogenetic inference
• point cloud
• random binning
• sampling maps
• sampling with replacement
• Tensors

• paper
• model release

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