James Petterson


The Bare Facts:

I'm a PhD student at the Research School of Information Science and Engineering of the Australian National University, and a research engineer at NICTA, where I am associated with the Statistical Machine Learning Program. Before shifting to an academic career I worked for 10 years as a software engineer in the telecommunications industry.


Contact:

Physical Address:
7 London Circuit, Tower A
Canberra ACT 2601
Australia
Postal Address:
Locked Bag 8001
Canberra ACT 2601
Australia
Phone:
(+61) 2 6267 6281
Email:
james.petterson@nicta.com.au

Publications:

[1]Exponential Family Graph Matching and Ranking.
NIPS, 2009. Accepted.
J. Petterson, T. Caetano, J. McAuley, J. Yu.
[pdf] [bib] [code]

[2]Distribution Matching for Transduction.
NIPS, 2009. Accepted.
N. Quadrianto, J. Petterson, A. Smola.
[pdf] [bib] 

[3]Hash Kernels for Structured Data.
JMLR, 2009. In press.
Q. Shi, J. Petterson, G. Dror, J. Langford, A. Smola, S. Vishwanathan.
[pdf] [bib] [code]

[4]Hash Kernels.
AISTATS, 2009.
Q. Shi, J. Petterson, G. Dror, J. Langford, A. Smola, A. Strehl, S. Vishwanathan.
[pdf] [bib] [code]

[5]Consistent Structured Estimation for Weighted Bipartite Matching.
NIPS Workshop on Algebraic and Combinatorial Methods in Machine Learning, 2009.
T. Caetano, J. Petterson, J. McAuley.
[pdf] [bib] [link] 

Software:

Large Scale Online Learning (Stream), a software platform for the implementation of large-scale online learning algorithms. This project is part of Elefant. You can check out the latest version of the source code with this command:
svn co http://elefant.developer.nicta.com.au/local/repos/trunk/elefant/stream
Parallel LDA collapsed Gibbs sampler in C++
I've been playing with parallelizing a LDA collapsed Gibbs sampler, and this code was my first attempt on that. It doesn't require any external libraries, is completely asynchronous and uses a shared directory to share information among CPUs. You can check the README for details, or see my results here. Even though the code was intended to be ran in parallel, it also works well running in a single CPU.

CV