|
Nonparametric supervised statistical learning neighborhood methods learn based on relative frequencies of samples that are 'near-neighbors' of a test point. We proposed and continue to explore the behavior of learning algorithms that use linear interpolation and some regularization strategy, such as the principle of maximum entropy. The first such approach is linear interpolation with maximum entropy (LIME). LIME weights have exponential form, the estimates are consistent, the estimates are robust to additive noise. The common linear interpolation solution used for regression on grids or look-up-tables has been shown to solve a related maximum entropy problem. LIME simulation results support use of the method, and performance on a pipeline integrity classification problem demonstrates that the proposed algorithm has practical value. Some of this work was with Stanford Statistician Richard Olshen and with Stanford Electrical Engineer Bob Gray. Publications: "Color Management of Printers by Regression over Enclosing Neighborhoods," Erika M. Chin, Eric K. Garcia and Maya R. Gupta , IEEE Intl. Conf. on Image Processing, 2007. "Minimum Expected Risk Estimation for Near-neighbor Classification," Maya R. Gupta, S. Srivastava and L. Cazzanti, Univ. of Washington Dept. of Electrical Engineering Technical Report 2006-0006, 2006. "Nonparametric supervised learning by linear interpolation with maximum entropy," Maya R. Gupta, Robert M. Gray, and Richard A. Olshen, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 5, pp. 766-781, 2006. LIME Code "On minimizing distortion and relative entropy," Michael P. Friedlander and Maya R. Gupta, IEEE Trans. on Information Theory, vol. 52, no. 1, pp. 238-245, 2006. "Minimum expected risk probability estimates for nonparametric neighborhood classifiers," Maya R. Gupta, Luca Cazzanti, and Santosh Srivastava, Proceedings of the IEEE Workshop on Statistical Signal Processing, 2005. "Quality assessment of low free-energy protein structure predictions," Luca Cazzanti, Maya Gupta, Lars Malmstrom, and David Baker, Proceedings of the IEEE Workshop on Machine Learning for Signal Processing, 375-380, 2005. "Inverting color transforms," Maya R. Gupta, Proceedings of the SPIE Electronic Imaging Conference on Computational Imaging II, vol. 5299, pp. 83-93, 2004. "An information theory approach to supervised learning," Maya R. Gupta, Ph.D. thesis (Dept. of Electrical Engineering, Stanford), 2003. "Analysis and classification of internal pipeline images," Deirdre O'Brien, Maya Gupta, Robert M. Gray, and Jon Kristian Hagene, Proceedings of the International Conference on Image Processing, pp. 577-580, 2003. "Automatic classification of images from internal optical inspection of gas pipelines," Deirdre O'Brien, Maya Gupta, Robert M. Gray, and Jon Kristian Hagene, Proceedings of the International Chemical and Petroleum Industry Inspection Technology VIII Conference, 2003. "Reducing bias in supervised learning," Maya R. Gupta and Robert M. Gray, Proceedings of the IEEE Workshop on Statistical Signal Processing, pp. 482-485, 2003. "Color conversions using maximum entropy estimation," Maya Gupta and Robert M. Gray, Proceedings of the International Conference on Image Processing, pp. 118-121, 2001. Related Work: |
|
 Back to Projects |
||