These files accompany the NIPS (December 2008, Vancouver) paper entitled "Diffeomorphic Dimensionality Reduction" by Christian Walder and Bernhard Schoelkopf. The files provide the result of our algorithm on the text data of the NIPS papers from Volumes 0 to 12, which we got from http://www.cs.toronto.edu/~roweis/data.html

We mapped the 1740 papers from their original number of words + number of authors = 13649 + 2037 = 15686 dimensional space to two dimensions. The mapped data are presented in the following files:

------------- FILE NUMBER ONE = diffeomap_nips_color.pdf 

This shows a the mapped points with a few hand selected authors color coded, to give an overview. Note that the color coding follows precisely that of the following paper:

Song, L., A. J. Smola, K. Borgwardt and A. Gretton: Colored Maximum Variance Unfolding. (NIPS 2007), MIT Press, Cambridge, MA.

------------- FILE NUMBER TWO = diffeomap_nips_titles.pdf

Here the mapped points are all plotted without color, but there is a markup in the pdf file which means that as you move the mouse over the dots, the paper title and author names should appear.

IMPORTANT: Some pdf readers do not work properly with this file. We had success with the latest adobe acrobat reader, which is free to download from www.adobe.com

------------- ALTERNATIVE TO FILE NUMBER TWO (ONLY IN CASE THERE IS A PROBLEM WITH THE PREVIOUS FILE)

In case there is a problem with the markup in the second pdf file, we offer a low tech alternative. In particular we provide a text file diffeomap_nips_list.txt which contains the results in the form

PAPER_NUMBER: (x,y) "TITLE" AUTHORS

where x and y correspond to the two dimensional space into which we mapped the data. In addition we provide diffeomap_nips_numbers.pdf which shows all of the paper numbers at their mapped locations, in a tiny font. By using both of these two files, diffeomap_nips_list.txt and diffeomap_nips_numbers.pdf, it is possible to browse the NIPS collection contextually. We recommend trying the free acrobat reader on the pdf files - this allows easy zooming and text searching, which together allow one to browse the the mapped data fairly easily. Once again, acrobat reader can be downloaded from www.adobe.com.

