PRISM is a plug-and-play meta-workflow for systematically constructing and evaluating time-series-to-image representations for multivariate anomaly detection. According to the abstract, more than 7,000 experiments compare PRISM configurations with 24 time-domain baselines. Selected configurations obtain the best VUS-PR on 10 of 14 datasets, averaging a 41% improvement over the strongest competing method on those datasets. The paper highlights channelization as a critical design choice and introduces MSM, a statistics-based scheme reporting 11–27% gains over PCA alternatives. Frozen ImageNet-pretrained encoders retain 92% of fine-tuned performance while training 1.8 times faster.
No heat snapshots are available in the last 24 hours.