Factored 3-Way Restricted Boltzmann Machines For Modeling Natural Images

By Marc'Aurelio Ranzato, Alex Krizhevsky, Geoffrey E. Hinton. 2010

252 citations recorded by the source corpus.

Subfields: Generative Models, Probabilistic & Bayesian Methods

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Generative Models

In Probabilistic & Bayesian Methods

Citations in the corpus

Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank (2013, background): “, 2012), extending Restricted Boltzmann machines (Ranzato and Hinton, 2010) and as a special layer for speech recognition (Yu et al.”
Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models (2014, methodology): “ear combination of slices weighted by each component u i of u. Here, the number of slices Gis equal to K, the dimensionality of u. It is often unnecessary to use a fully unfactored tensor. As in e.g. [38, 39], we re-represent T in terms of three matrices Wfk 2R FK, Wfd 2R G and Wfv 2RF V , such that”
On the Quantitative Analysis of Decoder-Based Generative Models (2016, methodology): “Following the implementation, we use SGD with momentum for optimization, and perform grid search of learning rate from {0.1, 0.5, 1, 2}, with momentum 0.9.”
Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks (2015, methodology): “Restricted Boltzmann machines [14, 18, 21, 23], Deep Boltzmann machines [26, 8], Denoising auto-encoders [30] all have a generative decoder that reconstructs the image from the latent representation.”
Predicting Parameters in Deep Learning (2013, methodology): “Our approach is superficially similar to the factored RBM [21, 26], whose parameters form a 3tensor.”
Representation Learning: A Review and New Perspectives (2012, methodology): “The PoT model has recently been generalized to the mPoT model (Ranzato et al., 2010b) to include nonzero Gaussian means by the addition of GRBM-like hidden units, similarly to how the mcRBM generalizes the cRBM.”
Generative Image Modeling Using Style and Structure Adversarial Networks (2016, background): “Some examples include the deep auto-encoder [22,23] or Restricted Boltzmann machines (RBMs) [24,25,26,27,28].”
A Multiplicative Model for Learning Distributed Text-Based Attribute Representations (2014, methodology): “Given an attribute vector x ∈ R D , we can compute attributed-gated word representations as T x = =1 x i T ( i ) i.e. word representations with respect to x are computed as a linear combination of slices weighted by each component x i of x .”
LRBM: A Restricted Boltzmann Machine Based Approach for Representation Learning on Linked Data (2014, background): “For simplicity, we first consider the case when weights of links are either 1 or 0, where 1 represents existence of interactions and 0 represents no interaction.”
On Autoencoders and Score Matching for Energy Based Models (2011, methodology): “Recently, more sophisticated latent variable EBMs for continuous data including the PoT (Welling et al., 2003), mPoT (Ranzato et al., 2010b), mcRBM (Ranzato & Hinton, 2010), FoE (Schmidt et al., 2010) and others have become popular models for learning representations of natural images as well as…”
Object recognition with hierarchical kernel descriptors (2011, unknown): “fine-tuning GRBM [24] 64.”
Kernel Descriptors for Visual Recognition (2010, background): “Deep belief networks have been extensively investigated on this dataset [21, 22].”
Efficient learning of sparse, distributed, convolutional feature representations for object recognition (2011, methodology): “2For example, L will be 128 when we use dense SIFT as an input.”
Independent component analysis: recent advances (2013, methodology): “A closely related formalism uses a generative model of the whole covariance structure of x [44,45].”
Gradient-based learning of higher-order image features (2011, methodology): “A common solution is to pre-project input images onto basisfunctions before computing pair-wise products [19], [25], [12].”
Gaussian sampling by local perturbations (2010, methodology): “We show that we can effectively replace the hybrid Monte Carlo sampler used by [18] with a block-Gibbs sampler in which the visible conditionally Gaussian units are sampled collectively by local perturbations, potentially allowing extension of the current patch-based model to a full-image factored G”
Winner-Take-All Autoencoders (2014, methodology): “or triangle k-means, by learning the connectivity between layers using a similarity metric on the feature maps, a ∼ 3% gain in classiﬁcation rate is achievable. Accuracy 3-Way Factored RBM (3 layers) [21] 65.3% Mean-covariance RBM (3 layers) [22] 71.0% Spike-and-Slab RBM [23] 76.7% Spike-and-Slab Sp”