Researchers propose TailSampler and TailedCore for unsupervised long-tail noisy anomaly detection.Existing models suffer from trade-off between tail class and noise samples.TailSampler estimates class size of tail samples to handle them separately.TailedCore outperforms state-of-the-art in unsupervised long-tail noisy anomaly detection.