Energy Based Large Margin Classification of Gaussian Mixture Model
Large margin GMMs have many parallels to large margin nearest neighbors (LMNN), but with classes modeled by ellipsoids instead of each input data and its target neighbors by hyper-ellipsoid. Large margin GMMs naturally scales to large problems in multi-way classification. Based on large margin GMM classification, we develop a new classification method, i.e., energy based large margin classification of Gaussian mixture mode (ELM-GMM). Experiment shows that this new approach outperforms the large margin GMMs.
C. G. Zhao "Energy Based Large Margin Classification of Gaussian Mixture Model", Advanced Materials Research, Vols. 271-273, pp. 1601-1604, 2011