Introduction
Summary: This page is a high level overview of some of the components in an active learning application. More details in following chapters!
A more complete high-level overview can be found in "NEXT: A System for Real-World Development, Evaluation, and Application of Active Learning".
NEXT is a system that connects algorithms with human responses. NEXT provides a framework that
- allows arbitrary algorithms (active or passive, etc) to gather human responses
- allows comparison of different algorithms easily
- allows for academic developers to use their algorithms (the original use case). Details like filenames and URLs are hidden from the algorithm developer by default (though not necessarily)
- allows monitoring of these experiments through "dashboards". These described in detail in Experiment-Monitoring dashboards provide access to
- algorithm output (rankings, embeddings, etc). These are shown per algorithm.
- participant responses. This details which objects were present on the screen, their ID, their response time, the network delay and more.
- timing information (how long did
getQuerytake for each algorithm?)
- provides a web user interface to collect human responses.
Talk given by Lalit Jain that explains high level features of NEXT to active learning researchers:
