Output details
11 - Computer Science and Informatics
University of Birmingham
Compressed Fisher Linear Discriminant Analysis: Classification of Randomly Projected Data
<24>Paper accepted for oral presentation in a top venue with acceptance rate 17% (13% oral). Proves that generalisation of a Fisher Linear Discriminant classifier can be achieved in randomly projected data spaces of dimension O(log #classes). This is a sharp improvement on O(log #points) previously known for margin based classifiers. Follow-on version won an IBM Best Student Paper Award at ICPR'10, with invited extension in Pattern Recognition Letters ICPR10 Awards Special Issue. It led to Kaban invited to speak at SIMPLE'12 workshop at MPI Dresden. Related ideas applied to optimisation (Kaban et al.) won a best paper award at GECCO'13.