Reproduce “BayPrAnoMeta: Bayesian Proto-MAML for Few-Shot Industrial Image Anomaly Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Soham Sarkar, Tanmay Sen, Sayantan Banerjee
BAYPRANOMETA introduces a federated Bayesian meta-learning framework for few-shot industrial image anomaly detection, integrating probabilistic normality models and supervised contrastive regularization to address data scarcity, uncertainty, and privacy constraints. The approach demonstrates improved anomaly detection performance and provides robust uncertainty estimates on the MVTec AD dataset.
Key insights
- The Normal-Inverse-Wishart prior ensures the Bayesian predictive model remains well-defined and numerically stable even when the support set size is smaller than the embedding dimension in few-shot scenarios.
- Supervised contrastive learning, which degrades performance in a centralized setup, proves beneficial in the federated setting by acting as a representation stabilizer, aligning features across heterogeneous clients.
- Probabilistic normality modeling with heavy-tailed Student-t distributions provides essential uncertainty quantification and robust anomaly scoring, outperforming deterministic prototype-based methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.