Guided Labeling - Generating Labels with Active Learning & Weak Supervision
Written by Paolo Tamagnini
One of the key challenges of utilizing supervised machine learning for real world use cases is that most algorithms and models require a lot of data and a number of specific requirements.
This means not only a sample of data large enough to represent the actual reality that the model needs to learn, but also labeled data. The problem here is that the labeling process is a tedious and drawn-out task and often expensive. So how can we efficiently improve this process and save on time and money? The answer is with two techniques: active learning and weak supervision.
In this book we discuss several of these two strategies and how to combine them. The solutions are implemented in KNIME workflows to be deployed to end-users as data apps. This results in a collection of data apps in which the subject matter expert is led through the active learning and/or weak supervision process of training a model.
This means not only a sample of data large enough to represent the actual reality that the model needs to learn, but also labeled data. The problem here is that the labeling process is a tedious and drawn-out task and often expensive. So how can we efficiently improve this process and save on time and money? The answer is with two techniques: active learning and weak supervision.
In this book we discuss several of these two strategies and how to combine them. The solutions are implemented in KNIME workflows to be deployed to end-users as data apps. This results in a collection of data apps in which the subject matter expert is led through the active learning and/or weak supervision process of training a model.