New framework calculates NML for non-smooth machine learning models

Researchers have developed a new theoretical framework for calculating the Normalized Maximum Likelihood (NML) for non-smooth models, which are common in modern machine learning. This approach uses geometric measure theory and automatic differentiation to ensure theoretical consistency. To implement this, they introduced a novel geometric MCMC algorithm called Propose-and-Project Metropolis-Hastings (PDL-PPMH), which can navigate non-differentiable level sets. The method was demonstrated to be a data-efficient alternative to cross-validation, achieving comparable predictive performance without needing to split data. AI

IMPACT Provides a more robust theoretical foundation for evaluating non-smooth machine learning models, potentially improving model selection and data efficiency.

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

Read on arXiv cs.LG →

• Lasso
• Propose-and-Project Metropolis-Hastings
• Sparse SVMs

• paper
• model release

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