- Abstract: We propose a new approach for using boosting to create an ensemble of generative models, where models are trained in sequence to correct earlier mistakes. Our algorithm can leverage many existing base learners, including recent latent variable models. Further, our approach allows the ensemble to leverage discriminative models trained to distinguish real data from model generated data. We show theoretical conditions under which incorporating a new model to the ensemble will improve the fit and empirically demonstrate the effectiveness of boosting on density estimation and sample generation on real and synthetic datasets.
- Conflicts: cs.stanford.edu
- Keywords: Theory, Deep learning, Unsupervised Learning