Microsoft Azure’s AI and machine learning offering Cognitive Services now contains a preview for an Anomaly Detector and so-called Custom Vision, a tool for building image classifiers.

Anomaly detection is something many companies use to either detect fraudulent behaviour or identify business incidents such as dropping rates in content providing scenarios in real time. The new Azure service is meant to let developers embed functions for detecting rare events or unusual patterns via an API to bump their application’s intelligence.

Custom Vision was available before but has only now reached general availability status.To get a model to recognise the image content relevant in a certain scenario, users will have to upload labeled images or use another service to tag untreated data.

The service will use those to generate a model, which can then seemingly be worked with as is or exported for offline usage on iOS or Android devices. For those under time constraints, Custom Vision offers an option to set a compute time budget, which the service uses as a basis to identify the best settings.

To further speed up machine learning in resource constrained environments, Microsoft also released a preview for hardware accelerated models to go with the Azure Data Box Edge appliance that has just been made available this week.

It is supposed to let developers make use of the known Azure Machine Learning services but packages them in Docker containers which can then be deployed to Microsoft’s new hardware via the Azure IoT Hub. Used in combination, the offering should lower latency and bandwidth costs, something many developers for edge use cases had to get creative with in the past.

As it stands, the preview can only be used to work with a variety of neural networks, namely ResNet-50, ResNet-152, DenseNet-121, and VGG16. According to the Data Box product pages, the monthly subscription fee for the Data Box Edge is $336.60 for now and provides a “cloud managed compute platform for containers at the edge”. To boost Azure cloud usage, the box also enables a data transfer to the cloud for storage or more complex analytics.

Another way to improve performance – though not machine learning related – is the newly announced Azure Stack HCI, which lets customers run virtualised applications on hyperconverged infrastructure (HCI). The offering is aimed at companies interested in going the hybrid cloud route, wanting to run workloads on premises as well as make use of the Azure management services for things like backup and disaster recovery. More details can be found in the introductory blog post.