Issue No. 08 - Aug. (2013 vol. 35)
R. Salakhutdinov , Dept. of Stat. & Comput. Sci., Univ. of Toronto, Toronto, ON, Canada
J. B. Tenenbaum , Dept. of Brain & Cognitive Sci., Massachusetts Inst. of Technol., Cambridge, MA, USA
A. Torralba , Comput. Sci. & Artificial Intell. Lab., Massachusetts Inst. of Technol., Cambridge, MA, USA
We introduce HD (or “Hierarchical-Deep”) models, a new compositional learning architecture that integrates deep learning models with structured hierarchical Bayesian (HB) models. Specifically, we show how we can learn a hierarchical Dirichlet process (HDP) prior over the activities of the top-level features in a deep Boltzmann machine (DBM). This compound HDP-DBM model learns to learn novel concepts from very few training example by learning low-level generic features, high-level features that capture correlations among low-level features, and a category hierarchy for sharing priors over the high-level features that are typical of different kinds of concepts. We present efficient learning and inference algorithms for the HDP-DBM model and show that it is able to learn new concepts from very few examples on CIFAR-100 object recognition, handwritten character recognition, and human motion capture datasets.
Approximation methods, Machine learning, Stochastic processes, Computational modeling, Vectors, Bayesian methods, Training,one-shot learning, Deep networks, deep Boltzmann machines, hierarchical Bayesian models
R. Salakhutdinov, J. B. Tenenbaum, A. Torralba, "Learning with Hierarchical-Deep Models", IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 35, no. , pp. 1958-1971, Aug. 2013, doi:10.1109/TPAMI.2012.269