Hi, I'm a graduate student at the University of Massachusetts Amherst in the Electrical and Computer Engineering Department. This semester, I'm involved with two semester long course projects (1) Bioinformatics and (2) Online Social Networks. Although having different application domains, both courses have common underlying themes--the problem of managing peta-scale data, and, using machine learning for sequence alignment/parameter inference (Bioinformatics), or, designing better recommender systems (Online Social Networks).
Having applied Markov Decision Process (MDP) to the area of CPU Power Management in the past, I think that modeling recommender systems as a sequential decision problem is a very interesting application of MDP [1]. Hidden Markov Models are commonly used in Gene Prediction/Sequence Alignment and can also be applied to Protein Structure determination problems. It is clear to me that Mahout has tremendous potential for application to diverse range of problems in the near future which involve Markov Models. However, the support for HMMs/MDPs is marginal (non-existent?) at the moment in Mahout. I would like to help in the development of Mahout and extend the framework, enabling it to handle HMM and or MDP. As I will be extending Mahout for the projects, I was wondering if there is any chance that my work could be used as a proposal to apply for the Google Summer of Code 2011 and build on the three months of development here at UMass. Any guidance in this regard is much appreciated. Thank you. Dhruv Kumar Electrical and Computer Engineering University of Massachusetts Amherst Amherst MA 01002 USA. [1] G. Shani et al. "An MDP-based recommender system." Journal of Machine Learning Research 2005.
