Paper Info
Title
Analysis-by-Synthesis by Learning to Invert Generative Black Boxes
Abstract
For learning meaningful representations of data, a rich source of prior knowledge may come in the form of a generative black box, e.g. a graphics program that generates realistic facial images. We consider the problem of learning the inverseof a given generative model from data. The problem is non-trivial because it is difficult to create labelled training cases by hand, and the generative mapping is a black box in the sense that there is no analytic expression for its gradient. We describe a way of training a feedforward neural network that starts with just one labelled training example and uses the generative black box to "breed" more training data. As learning proceeds, the training set evolves and the labels that the network assigns to unlabelled training data converge to their correct values. We demonstrate our approach by learning to invert a generative model of eyes and an active appearance model of faces.
Year
DOI
Venue
2008
10.1007/978-3-540-87536-9_99
ICANN (1)
Keywords
Field
DocType
training data,generative mapping,active appearance model,training set evolves,labelled training case,unlabelled training data,generative model,invert generative black boxes,black box,labelled training example,generative black box,feedforward neural network,analysis by synthesis,graphical programming
Black box (phreaking),Generative topographic map,Computer science,Generative Modelling Language,Active appearance model,Artificial intelligence,Generative grammar,Black box,Machine learning,Generative Design,Generative model
Conference
Volume
ISSN
Citations 
5163
0302-9743
11
PageRank 
References 
Authors
0.89
6
3
Name
Order
Citations
PageRank
Vinod Nair11658134.40
Josh Susskind2110.89
geoffrey e hinton3404354751.69