Active Learning Classifier
What makes this innovative?
The concept and solution is widely applicable, easily adjustable and helps to automate a wide range of processes. It builds on the existing workflows and provides a natural transition from manual tasks to full automation, saving costs, resources and frustration that comes from repetitive tasks.
Data-based manual tasks often bind a lot of resources from domain experts
Transition from manual tasks to full automation free up expert resources
Clients often face the challenge of having data-based manual tasks that bind a lot of resources and are not the most exciting for the employees. The tasks range from sorting emails, tagging documents, mapping elements in different schemas to evaluating and classifying images. The tasks often require domain knowledge and need to be done by experts.
The solution we build is a flexible active learning workflow. Active learning helps to build and optimize a machine learning model while assisting the expert doing the task. The model gradually becomes better until the point that the expert trusts it to take over the task completely.
Our solution is build around a flexible classification pipeline that can be customized for various data types and use-cases. The model is embedded in a production ready API that can be quickly deployed. We also added a simple to use web-app to try the active learning workflow in practice.
The active learning classifier helps to (semi-) automate processes and free up expert resources. Having a trained model also provides more consistent results than different experts working the same task.
FRAMEWORK & TOOLS
Python, FastAPI, Sklearn