New framework generates structured outliers to test ML model robustness

Researchers have introduced Sinkhorn Boundary Outlier Generation (SBOG), a novel framework designed to create structured outliers for evaluating and enhancing machine learning model robustness. This method utilizes Sinkhorn optimal transport geometry and distributionally robust boundary modeling to guide the generation process. SBOG aims to produce semantically controlled deviations from in-distribution data, moving beyond traditional heuristic approaches that rely on sparse neighborhoods or classifier boundaries. Experiments demonstrate SBOG's effectiveness in generating informative outliers for time series anomaly detection and image synthesis, ultimately improving downstream robustness evaluations. AI

IMPACT Enhances ML model robustness testing by providing a structured method for generating informative outliers beyond empirical data.

RANK_REASON The cluster contains a research paper detailing a new method for generating outliers to test machine learning robustness. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• ScienceCast
• Sinkhorn Boundary Outlier Generation
• Sinkhorn Optimal Transport

• paper
• safety

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