Researchers have released a preprint describing NV-Tesseract-AD, a framework for multivariate time-series anomaly detection. The method builds on conditional diffusion models and adds a curriculum-learning component, where training tasks are ordered by difficulty. The authors' central hypothesis is that reconstruction difficulty can be explicitly structured during training, rather than left implicit.
Because the abstract is cut off, the full details of how the curriculum is designed or evaluated are not available from the source. The paper is listed on arXiv under identifier 2610.12754.