Unsupervised Learning of Image Transformations

By Roland Memisevic, Geoffrey E. Hinton. 2007

217 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning

How researchers see this paper

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

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models (2014, methodology): “As in e.g. [38, 39], we re-represent T in terms of three matrices where diag ( · ) denotes the matrix with its argument on the diagonal.”
Learning to generate chairs with convolutional neural networks (2014, background): “RBMs and DBMs are undirected graphical models which aim to build a probabilistic model of the data and treat encoding and generation as an (intractable) joint inference problem.”
Attribute2Image: Conditional Image Generation from Visual Attributes (2015, background): “As a representative example, the restricted Boltzmann machine (RBM) [33] and its variants [28, 25, 30, 15, 37, 31] have been widely studied for image modeling.”
RGB-D-based Human Motion Recognition with Deep Learning: A Survey (2017, neutral): “Taylor et al. [83] extended the Gated RBM (GRBM) [84] to convolutional GRBM (convGRBM) that shares weights at all locations in an image and inference is performed through convolution.”
Perspective Transformer Nets: Learning Single-View 3D Object Reconstruction without 3D Supervision (2016, background): “Memisevic and Hinton [12] introduced a gated Boltzmann machine that models the transformations between image pairs using multiplicative interaction.”
Multi-view 3D Models from Single Images with a Convolutional Network (2015, background): “These often involve multiplicative interactions, for example gated RBMs [2], gated autoencoder [3] or Disentangling Boltzmann Machines [4].”
Semantic Compositional Networks for Visual Captioning (2016, methodology): “In order to remedy this problem, we adopt ideas from [30] to factorize W(s) and U(s) defined in (5) as”
Learning to Relate Images (2011, methodology): “Like an auto-encoder, this model can be trained with gradient based optimization.”
A Multiplicative Model for Learning Distributed Text-Based Attribute Representations (2014, background): “The LBL model makes a linear prediction of the next word representation as r̂ = n−1∑ i=1 C(i)rwi , (1) where C(i), i = 1, . . . , n − 1 are K × K context parameter matrices.”
Learning to Represent Spatial Transformations with Factored Higher-Order Boltzmann Machines (2010, methodology): “Alternatively, the model can be viewed as an RBM whose connections are modulated by input pixels, which can consequently “vote” for linear filters that jointly model the output image (Memisevic and Hinton, 2007).”
Visual Prompting via Image Inpainting (2022, background): “Predicting the correct completion was previously modeled as a classiﬁcation problem [43, 23], and other works have explored analogies in the context of learning different transformations between pairs of images [34, 46].”
Pose-Sensitive Embedding by Nonlinear NCA Regression (2010, background): “Alternatively, temporal context could be integrated at the input level, from simple frame differencing to more sophisticated temporal feature extraction (e.g. [23]).”
Building a better probabilistic model of images by factorization (2011, background): “Previous work on modeling image transformations [21, 25, 22], separating content and style [7, 10, 11], and capturing higher-order dependencies in natural images [14, 15, 27], draws upon a simple but powerful bilinear model, in which an observation x is described by a mixture of basis elements with ”
Unsupervised Learning of Visual Structure using Predictive Generative Networks (2015, background): “Also related to our approach, especially in the context of rotating objects, is the ﬁeld of relational feature learning (Memisevic & Hinton, 2007; Taylor & Hinton, 2009).”
Bimodal Modelling of Source Code and Natural Language (2015, methodology): “While the gating behavior of multiplicative models has been discussed previously (Memisevic & Hinton, 2007; Taylor & Hinton, 2009; Kiros et al., 2013), our aim here is to explain the importance of these multiplicative interactions in the context of source code modelling, and to point out a concrete…”
Bootstrapping bilinear models of Simple Vehicles (2015, methodology): “There is also a computational similarity with deep learning methods using three-way interactions (Memisevic and Hinton, 2007; Larochelle and Hinton, 2010; Memisevic and Hinton, 2010; Ranzato and Hinton, 2010), as well as subspace approaches in the system identification literature for learning…”
A Review of Deep Learning with Special Emphasis on Architectures, Applications and Recent Trends (2019, methodology): “Temporal RBM [57] and conditional RBM [58] proposed and applied to model multivariate time series data and to generate motion captures, Gated RBM [59] to learn transformation between two input images, Convolutional RBM [60], [61] to understand the time structure of the input time series,…”