DRAW: A Recurrent Neural Network For Image Generation (2015-02-16T00:00:00.000000Z)

TL;DR

The Deep Recurrent Attentive Writer neural network architecture for image generation substantially improves on the state of the art for generative models on MNIST, and, when trained on the Street View House Numbers dataset, it generates images that cannot be distinguished from real data with the naked eye.

Abstract

This paper introduces the Deep Recurrent Attentive Writer (DRAW) neural network architecture for image generation. DRAW networks combine a novel spatial attention mechanism that mimics the foveation of the human eye, with a sequential variational auto-encoding framework that allows for the iterative construction of complex images. The system substantially improves on the state of the art for generative models on MNIST, and, when trained on the Street View House Numbers dataset, it generates images that cannot be distinguished from real data with the naked eye.

Authors

Daan Wierstra

12 papers

Alex Graves

11 papers

Danilo Jimenez Rezende

5 papers

TL;DR

Abstract

Authors

References36 items

A Neural Autoregressive Approach to Attention-based Recognition

Multiple Object Recognition with Visual Attention

Attention for Fine-Grained Categorization

Adam: A Method for Stochastic Optimization

Optimizing Neural Networks that Generate Iimages

Neural Turing Machines

Sequence to Sequence Learning with Neural Networks

Recurrent Models of Visual Attention

Iterative Neural Autoregressive Distribution Estimator NADE-k

Neural Variational Inference and Learning in Belief Networks

Stochastic Backpropagation and Approximate Inference in Deep Generative Models

Learning Generative Models with Visual Attention

Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks

Auto-Encoding Variational Bayes

Deep AutoRegressive Networks

A Deep and Tractable Density Estimator

Generating Sequences With Recurrent Neural Networks

Learning Where to Attend with Deep Architectures for Image Tracking

Learning to combine foveal glimpses with a third-order Boltzmann machine

Deep Boltzmann Machines

Evaluating probabilities under high-dimensional latent variable models

On the quantitative analysis of deep belief networks

Learning to Forget: Continual Prediction with LSTM

Learning where to look

Long Short-Term Memory

The Helmholtz Machine

Markov Chain Monte Carlo and Variational Inference: Bridging the Gap

Reading Digits in Natural Images with Unsupervised Feature Learning

The Neural Autoregressive Distribution Estimator

Learning Multiple Layers of Features from Tiny Images

Supporting Online Material for Reducing the Dimensionality of Data with Neural Networks

Gradient-based learning applied to document recognition

Experiments – Generating CIFAR Images

Experiments – Cluttered MNIST Classification Experiments – MNIST Generation

Experiments – MNIST Generation with Two Digits 13 / 16

Experiments – Street View House Number Generation 14 / 16

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names