Constrained Data Compression

Inserat Nr.

Montag, 9. Januar 2017

ETH Zürich


Nach Vereinbarung


Constrained data compression

Dimensionality reduction is a very useful technique for visualizing high dimensional data. Recently here has been much progress in dimensionality reduction, as exemplified by t-SNE, which is a technique that is able to capture meaningful clusters in complex high-dimensional data sets (van der Maaten and Hinton, 2008).

Dimensionality reduction has not been worked out well in cases in which not all data samples are equally trustworthy or in which subsets of data samples are bound together via diverse constraints. Currently, there is no method that allows combined data constraining and compression. Such problems frequently occur in unsupervised problems of joint visual segmentation and classification.

We propose an exciting research project in statistical learning. The goal is to explore theoretical frameworks for combining techniques in constrained optimization and dimensionality reduction. 

Studennts interested  in a Semster/Bachelor/Master Thesis please get in touch with Richard Hahnloser, hrichard AT


L.J.P. van der Maaten and G.E. Hinton. Visualizing High-Dimensional Data Using t-SNE. Journal of Machine Learning Research 9(Nov):2579-2605, 2008.