Active Learning as a Service

For many companies, there is a huge cost in cleaning and labeling data for AI applications. However, datas are not equal, some data samples contain more information which means labeling those data can get more benefits. Active learning is a method to select the most informative data samples for labeling to achieve a comparable performance with labbeling all the data, and reduces the labeling budget. However, active learning has some downsides, 1. active laerning is hard to implement in the real-world, it requires additional engineering works to build a practical pipeline. 2. active learning runs slowly, and cost extra conputational resources for selecting the data and updating the learning model. Our ALaaS trys to bridge the gaps, we made active learning as a service, which can be deployed in a fast and convient way. And the built-in designs in ALaaS can help to reduce the resource usages of active learning, make it more efficient. The project is still under active developments, welcome to discuss and make contributions. Thank you!