Back from ICML …

June 25, 2007 7:02 pm by Markus.

Just got back in town from ICML 2007, had a blast and met lots of old friends. This year it felt a bit more like a camping trip with no hot water and filthy bathrooms. Otherwise I learned a lot, specifically the following papers were in my opinion the most interesting (in no particular order):

• Pegasos: Primal Estimated sub-GrAdient SOlver for SVM (Shai Shalev-Shwartz, Yoram Singer, Nathan Srebro):a very fast online SVM with good bounds, kernelizable. Code available. Most impressive results and probably useful for the robot stuff we are working on.
• A Kernel Path Algorithm for Support Vector Machines (Gang Wang, Dit-Yang Yeung, Frederick Lochovsky). Speed up SVM learning by not having to re-train when the Kernel Sigma is changed. I hope they will make some code available
• Restricted Boltzmann Machines for Collaborative Filtering (Ruslan Salakhutdinov, Andriy Mnih, Geoffrey Hinton). Now #4 at the Netflix Challenge. I already wrote in my AISTATS post that I think this technique has a lot of potential.
• Multi-Armed Bandit Problems with Dependant Arms (Sandeep Pandey, Deepayan Chakrabarti, Deepak Agarwal) A Clustering trick to distribute rewards and speedup reinforcement learning in instances of banner advertisings
• CarpeDiem: an Algorithm for the Fast Evaluation of SSL Classifiers (Roberto Esposito, Daniele Radicioni) A fast Viterbi
• Graph Clustering with Network Structure Indices (Matthew Rattigan, Marc Maier, David Jensen) Fast, simple graph-mining algorithms. Since I’m currently reading “Mining Graph Data”….
• A Dependance Maximization View of Clustering (Le Song, Alex Smola, Arthur Gretton, Karsten Borgwardt) An interesting, general framework for Clustering using the Hilbert-Schmidt Independence Criterion that makes many clustering algorithms (K-Means, Spectral Clustering, Hierarchical Clustering) mere special cases…
• Neighbor Search with Global Geometry: A Minimax Message Passing Algorithm (Kye-Hyeon Kim, Seungjin Choi). Interesting idea…

I just notice that my paper-list is exceptionally long this time. So I did get a lot of cool new things out of it; I’m glad I went. I will hopefully be able to try some of my ideas soon(mostly busy with thesis writing right now)…

Posted in Machine Learning | Print | No Comments »