New Bayesian header improves Vision Transformers' robustness to noisy labels

Researchers have developed a new Bayesian header, termed LipB-ViT, designed to improve the robustness of vision transformers against label noise. This architecture-agnostic header enforces spectral normalization on variational weights, leading to better calibrated uncertainty and reduced noise amplification. The method also introduces novel metrics for assessing dataset quality and label noise, outperforming existing techniques in detecting semantically misclassified labels. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel method to improve model robustness against label noise, potentially enhancing reliability in high-stakes applications with variable annotation quality.

RANK_REASON This is a research paper detailing a novel method for improving model robustness against label noise. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• LipB-ViT
• arXiv
• vision transformers
• Bayesian header
• spectral normalization
• Monte Carlo sampling

• paper
• safety