Changes between Version 18 and Version 19 of MatrixFactorization
 Timestamp:
 07/21/11 15:15:01 (3 years ago)
Legend:
 Unmodified
 Added
 Removed
 Modified

MatrixFactorization
v18 v19 22 22 23 23 === nsnmf === 24 Nonsmooth NMF. Uses a modif ed version of Lee and Seung's multiplicative updates for KullbachLeibler divergence. It is meant to give sparser results.[[BR]]24 Nonsmooth NMF. Uses a modified version of Lee and Seung's multiplicative updates for KullbachLeibler divergence. It is meant to give sparser results.[[BR]] 25 25 Reference: (PascualMontano, 2006). 26 26 … … 34 34 35 35 === pmf === 36 Probabilistic MF. PMF model scales linearly with the number of observations and performs well on large, sparse, imbalanced datasets.[[BR]] 37 Reference: (Salakhutdinov, Mnih 2008 (NIPS)). 38 39 === Additional: bpmf === 40 Bayesian PMF model that implements Gibbs sampler. In Bayesian PMF model capacity is controlled automatically by integrating over all model parameters and hyperparameters. Bayesian PMF can be efficiently trained using MCMC methods. This model achieves significantly higher prediction accuracy than PMF models trained using MAP estimation. This algorithm has been used by winning and top ranked teams in KDD Cup 2011. [[BR]] 41 Reference: (Salakhutdinov, Mnih, 2008 (ICML)). 36 Probabilistic MF. Target matrix is interpreted as samples from multinomial. [[BR]] 37 Reference: (Hansen, 2008), (Laurberg, 2008). 42 38 43 39 === psmf === … … 145 141 * Zhang, Z., Li, T., Ding, C. H. Q., Zhang, X. Binary Matrix Factorization with Applications. ICDM 2007. 146 142 * Salakhutdinov, R., Mnih, A. Bayesian probabilistic matrix factorization using Markov chain Monte Carlo. ICML 2008, 880887. 143 * Laurberg, H.,et. al. Theorems on positive data: on the uniqueness of NMF. Computational intelligence and neuroscience. 2008. 144 * Hansen, L. K. Generalization in highdimensional factor models. 2008. Web: http://www.stanford.edu/group/mmds/slides2008/hansen.pdf.