[ad_1]
You can not create a good machine learning model without good training data. But building these training sets is a difficult job, often manual, which involves labeling thousands and thousands of images, for example. With SageMaker, AWS is working on a service that makes it much easier to build machine learning models. But until today, this labeling task still rested with the user. Now, however, the company is launching SageMaker Ground Truth, a labeling service for training packages.
With the help of Ground Truth, developers can direct the service to the storage bins containing the data and allow it to tag them automatically. What is interesting here is that you can set a level of trust for fully automatic service or send data to workers. These human labellers, who probably occupy the most scathing position in the technology sector, may be Mechanical Turk users or third-party services. If you really hate your employees, you can also ask them to tag.
Currently, the service supports text clbadification, image clbadification, object detection, and semantic segmentation. Users can also create their own tasks.
As the labeling data arrives, Ground Truth extracts some of the objects and sends them to the human labellers to create a new custom template for the user.
"We are able to tag your data," said Andy Jbady, CEO of AWS. "So you can build those types of models that were previously difficult or impossible or too expensive to produce."
Source link
