PDE-Foam
The PDE-Foam method is an extension of the PDE-RS algorithm used for multi-variate probability density estimation. It divides a multi-dimensional phase space in a finite number of hyper-rectangles (cells) of constant event density. This "foam" of cells is filled with averaged probability-density information sampled from a training event sample. For a given number of cells, the binning algorithm adjusts the size and position of the cells inside the multidimensional phase space based on a binary-split algorithm, minimizing the variance of the event density in the cell. The binned event density information of the final foam is stored in cells, organised in a binary tree, allowing for a fast and memory-efficient storage and retrieval of the event density information necessary for the classification of events.
PDE-Foam is part of TMVA since version 4.0.0.
CERN Summer Student Project 2008
CERN project report
Talk at IKTP institute seminar
Paper preprint