The main goal of machine learning (ML) is to design, study and apply algorithms for learning from past experience, and is particular useful as soon as interaction becomes amenable for automation. Therefore it balances between computational statistics, computer sciences, numerical algebra, mathematics and signal processing. It studies a toolbox of methods for assisting the human expert in decision making. The usual approach of ML is to process historical data into useful predictions about future cases.
Machine Learning techniques can generally be divided in the supervised and the unsupervised techniques: the first try to uncover relations in the data for a specific (labeled) phenomenon, while the latter looks for consistent patterns without any guidance from labeled data. In the context of VASCO, we’re essentially interested in an unsupervised technique as we don’t want to specify what exactly we’re after: phenomenon that differ between the sky surveys do not necessarily have a known explanation as yet. This brings us in the domain of pattern recognition, clustering and visualisation.
There are ample of good starter kits on the web, but the ones we use are
- scikit-learn (a python toolbox for machine learning) (www).
- The elements of statistical learning (springer) (PDF), or the easier PDF.
- The massively popular ML course by Andrew Ng.