Automatic identification of metaparameters in machine learning and signal processing applications (with UFSC)

This project will focus on the methods and algorithms which facilitate automatic choice of the parameters required in estimation/learning from data. We will focus on real-time algorithms such as those appearing in adaptive signal processing and on-line learning. The project will be particularly focus on the identification of meta-parameters in real-time from the data while performing the estimation/learning. This is a well-known, albeit not yet solved, problem in signal processing, where meta parameters are often heuristically defined. The same issue appears in machine learning when parameters defining the convergence or regularization have to be decided prior to the estimation. We will study methods which find/estimate these parameters at the same time as the learning takes place and should become a valuable tool to fully automatize the adaptive algorithms.

Faculty Supervisor:

Leszek Szczecinski

Student:

Partner:

Universidade Federal de Santa Catarina

Discipline:

Engineering

Sector:

Education

University:

Université du Québec : Institut national de la recherche scientifique

Program: