IBM Research Post Details Real-Time Anomaly Detection in Confluent Flink SQL; No Accuracy Benchmarks
Reported by Tech Times1 source ↓
IBM Research describes calling Granite time-series models from Confluent Flink SQL to forecast streaming data and flag anomalies. Confluent marks the cloud functions Early Access and the on-premises version Open Preview; IBM reports no accuracy or latency benchmarks.
Key points
- Confluent Cloud labels AI_FORECAST and AI_DETECT_ANOMALIES Early Access for evaluation and non-production testing
- Confluent Platform for Apache Flink lists the functions as Open Preview, and says interfaces may change
- AI_DETECT_ANOMALIES flags readings outside forecast prediction bounds; its default confidence setting is 99%
- The function waits for 20 data points before producing output and retains up to 512 recent values, depending on model
- IBM's post reports no accuracy or latency benchmarks and names no customer deployments
Details
Tech Times reports that IBM's Nov. 18 Research post describes using Granite time-series models through Confluent Flink SQL functions in both Confluent Cloud and Confluent Platform, which customers operate on their own infrastructure. The post presents six example scenarios, including screening card payments for fraud, but does not report accuracy or latency results or name customer deployments.
Confluent's documentation says AI_DETECT_ANOMALIES uses a model forecast and prediction intervals: readings below the lower bound or above the upper bound are flagged. Confidence defaults to 99%, and the function's output includes the forecast, bounds, anomaly flag and a root-mean-square error figure. It produces no output until it has seen 20 points and retains up to 512 recent values, a limit that varies by model.
Confluent lists TTM, FlowState and PatchTST-FM as selectable models, with TTM as the default. Its cloud documentation labels AI_FORECAST and AI_DETECT_ANOMALIES Early Access features for evaluation and non-production testing, while its on-premises release post labels the functions Open Preview and warns that interfaces may change.
Written from coverage as of
Images
- Tech Times
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh