Slow Feature Analysis for Unsupervised and Supervised Learning
Staff - Faculty of Informatics
Start date: 18 December 2014
End date: 19 December 2014
The Faculty of Informatics is pleased to announce a seminar given by Laurenz Wiskott
DATE: Thursday, December 18th, 2014
PLACE: USI Lugano Campus, room SI-003, Informatics building (Via G. Buffi 13)
TIME: 13.30
ABSTRACT:
Slow feature analysis (SFA) is a biologically motivated algorithm for extracting slowly varying features from a quickly varying signal and has proven to be a powerful general-purpose preprocessing method for spatio-temporal data. We have applied SFA to the learning of complex cell receptive fields, visual invariances for whole objects, and place cells in the hippocampus. On the technical side SFA can be used to extract slowly varying driving forces of dynamical systems and to perform nonlinear blind source separation.
More recently we have developed methods to generalize SFA to supervised learning of data without explicit time structure but high dimensional input vectors, in particular for face processing. The basic idea is to convert unstructured but labeled data into structured data without labels suitable for unsupervised learning methods, then perform hierarchical processing to reduce the dimensionality, and finally perform supervised learning on the reduced data.
In this talk I will give an overview over these applications with a focus on the self-organization of hippocampal place cells and supervised learning on face images.
BIO:
Laurenz Wiskott studied physics in Göttingen and Osnabrück, Germany, and received his PhD in 1995 at the Ruhr-University Bochum. The stages of his career include three-years at The Salk Institute in San Diego, one year at the Institute for Advanced Studies in Berlin, and nine years at the Institute for Theoretical Biology, Humboldt-University Berlin, where he was heading a junior research group and became Professor in 2006. Since 2008 he is at the Institute for Neural Computation, Ruhr-University Bochum. He has been working in the fields of Computer Vision, Neural Networks, Machine Learning and Computational Neuroscience.
HOST: Prof. Jürgen Schmidhuber
