Microsoft has emitted a couple of technologies for its Azure Cognitive Services designed to spot unusual patterns and classify images.
First up is a preview of a new Cognitive Service, Anomaly Detector. With "does what it says on the tin" branding, the tech is built to detect unusual patterns or rare events in data.
While Microsoft is keen to point out the tech is already in use across the company to identify irregularities with a view to speeding up troubleshooting, other use cases include monitoring big data flows, such as IoT traffic, keeping an eye on streaming video traffic and, of course, dealing with possible transactional naughtiness.
The latter will be of particular interest to the financial sector. An industry insider told us that current technologies, such as Oracle's Mantas Financial Services Anti Money Laundering were, at their heart, scripts running scenarios on the huge data sets involved.
As an example, a scenario would look for transfers over £10k, with banks shifting that threshold up and down should miscreants get wise to the rules. The results can then swamp backroom teams as ops personnel check flagged transactions. Our insider told us: "Some may be false positives," before adding, ruefully, "Most are."
Applying a bit of machine learning wizardry to such things is therefore key in improving the quality of the results (and most likely swinging an axe on the necks of those charged with checking the output).
Microsoft, of course, is not the only game in town. HSBC has buddied up with Google Cloud for a project focused on machine learning to spot anomalies. The quantities of data involved meant a partnership with one of the big cloud providers, like Google, proved inevitable.
Or Microsoft, who has invested big in Azure's AI smarts.
The timing for the new service is good. Our insider remarked there is a push away from traditional rules-based transaction monitoring towards machine learning, in the hope of improving the quality of alerts. He noted, however, that things were still a long way off.
The all-seeing eye of Custom Vision
While the Anomaly Detector service remains in preview, the Azure gang has pushed Custom Vision, Microsoft's take on identifying objects in images, to general availability.
The tech allows developers to train their own classifier to flag what they consider important and, handily, export the things to be used offline (on iOS, Android and, of course, edge devices). The exported models are tweaked to handle the limitations of mobile devices.
Microsoft cited the use of the technology to detect foam levels in water being treated for livestock as an example in the field.
While image classification is nothing new (TensorFlow and its like have been doing it for years), the simplicity of Custom Vision is appealing from a developer perspective. SDKs for .NET, Python, Java, Node.js and Go mean most tastes are catered for, and we were able to run up an app to look for logos in images with little difficulty.
Classifiers can also be exported to be used by Raspberry Pi 3 computers. Their edgier ilk will get a look-in via the upcoming Vision AI Developer Kit. ®