Back
Deterministic annealing for semi-supervised kernel machines
An intuitive approach to utilizing unlabeled data in kernel-based classification algorithms is to simply treat the unknown labels as additional optimization variables. For margin-based loss functions, one can view this approach as attempting to learn low-density separators. However, this is a hard optimization problem to solve in typical semi-supervised settings where unlabeled data is abundant. The popular Transductive SVM algorithm is a label-switching-retraining procedure that is known to be susceptible to local minima. In this paper, we present a global optimization framework for semi-supervised Kernel machines where an easier problem is parametrically deformed to the original hard problem and minimizers are smoothly tracked. Our approach is motivated from deterministic annealing techniques and involves a sequence of convex optimization problems that are exactly and efficiently solved. We present empirical results on several synthetic and real world datasets that demonstrate the effectiveness of our approach.
@inproceedings{4061, title = {Deterministic annealing for semi-supervised kernel machines}, journal = {Proceedings of the 23rd International Conference on Machine Learning (ICML 2006)}, booktitle = {ICML 2006}, abstract = {An intuitive approach to utilizing unlabeled data in kernel-based classification algorithms is to simply treat the unknown labels as additional optimization variables. For margin-based loss functions, one can view this approach as attempting to learn low-density separators. However, this is a hard optimization problem to solve in typical semi-supervised settings where unlabeled data is abundant. The popular Transductive SVM algorithm is a label-switching-retraining procedure that is known to be susceptible to local minima. In this paper, we present a global optimization framework for semi-supervised Kernel machines where an easier problem is parametrically deformed to the original hard problem and minimizers are smoothly tracked. Our approach is motivated from deterministic annealing techniques and involves a sequence of convex optimization problems that are exactly and efficiently solved. We present empirical results on several synthetic and real world datasets that demonstrate the effectiveness of our approach.}, pages = {841-848}, editors = {Cohen, W. W., A. Moore}, publisher = {ACM Press}, organization = {Max-Planck-Gesellschaft}, school = {Biologische Kybernetik}, address = {New York, NY, USA}, month = jun, year = {2006}, slug = {4061}, author = {Sindhwani, V. and Keerthi, SS. and Chapelle, O.}, month_numeric = {6} }