New KReTTaH framework offers data-free imputation via tensor trains

A new framework called KReTTaH has been introduced for multi-way data imputation, utilizing kernel regression with tensor trains and Hadamard overparameterization. This method is designed to be training-data-free, interpretable, and nonparametric. KReTTaH reformulates the imputation problem as regression within reproducing kernel Hilbert spaces, constraining tensor regression coefficients to fixed-rank tensor-train manifolds and employing Hadamard overparameterization for efficiency. Numerical tests on functional magnetic resonance imaging data and dynamic graph recovery show KReTTaH surpasses current state-of-the-art tensor, Bayesian, and neural network methods in accuracy. AI

IMPACT This framework could improve data imputation accuracy in complex datasets, potentially benefiting AI models that rely on complete data for training and inference.

RANK_REASON The cluster contains a research paper detailing a new statistical framework and its application. [lever_c_demoted from research: ic=1 ai=0.7]

Read on arXiv stat.ML →

• arXiv
• Auguste Hadamard
• functional magnetic resonance imaging
• Hugging Face
• KReTTaH
• Reproducing Kernel Hilbert Spaces
• Tensor-Train Decomposition

• paper
• model release

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