Recurrent neural networks for data item generation

Patent metadata

Field	Value
Publication number	US-11080587-B2
Application number	US-201615016160-A
Country	US
Kind code	B2
Filing date	Feb 4, 2016
Priority date	Feb 6, 2015
Publication date	Aug 3, 2021
Grant date	Aug 3, 2021

Abstract

Methods, and systems, including computer programs encoded on computer storage media for generating data items. A method includes reading a glimpse from a data item using a decoder hidden state vector of a decoder for a preceding time step, providing, as input to a encoder, the glimpse and decoder hidden state vector for the preceding time step for processing, receiving, as output from the encoder, a generated encoder hidden state vector for the time step, generating a decoder input from the generated encoder hidden state vector, providing the decoder input to the decoder for processing, receiving, as output from the decoder, a generated a decoder hidden state vector for the time step, generating a neural network output update from the decoder hidden state vector for the time step, and combining the neural network output update with a current neural network output to generate an updated neural network output.

First claim

What is claimed is:

1. A neural network system implemented by one or more computers, the neural network system comprising:
an encoder neural network, wherein the encoder neural network is a recurrent neural network that is configured to, for each input image processed by the encoder neural network and at each time step of a plurality steps:
receive a glimpse captured by reading from the input image,
receive a decoder hidden state vector of a decoder neural network for the preceding time step, and
process the glimpse, the decoder hidden state vector, and an encoder hidden state vector of the encoder neural network from the preceding time step to generate an encoder hidden state vector for the time step;

a decoder neural network, wherein the decoder neural network is a recurrent neural network that is configured to, for each of the plurality of time steps:
receive a decoder input for the time step, and
process the decoder hidden state vector for the preceding time step and the decoder input to generate a decoder hidden state vector for the time step; and

a subsystem, wherein the subsystem is configured to generate a final output image by repeatedly updating a neural network output at each of a plurality of time steps based on an input image to generate a final neural network output and generating the final output image from the final neural network output, the updating comprising, for each of the time steps:
reading the glimpse from the input image using the decoder hidden state vector for the preceding time step;
providing the glimpse as input to the encoder neural network;
generating the encoder hidden state vector at the time step as output from the encoder neural network;
generating the decoder input for the decoder neural network from the encoder hidden state vector at the time step;
providing the decoder input as input to the decoder neural network for the time step;
generating a neural network output update for the time step from the decoder hidden state vector for the time step; and
combining the neural network output update for the time step with a current neural network output to generate an updated neural network output.

2. The neural network system of claim 1 , wherein the encoder neural network and the decoder neural network are long short term memory neural networks.

3. The neural network system of claim 1 , wherein the subsystem is further configured to train the encoder neural network and the decoder neural network to autoencode input images.

4. The neural network system of claim 3 , wherein training the encoder neural network and the decoder neural network to autoencode input images comprises training the neural networks to generate a neural network output image that is a reconstruction of the input image.

5. The neural network system of claim 1 , wherein the subsystem is further configured to provide the encoder hidden state vectors from each of the time steps for a particular image as features of the particular image.

6. The neural network system of claim 5 , wherein the features of the particular image are provided for use in processing the particular image during a semi-supervising learning procedure.

7. The neural network system of claim 1 , wherein in the input data items are image frames from videos.

8. The neural network system of claim 1 , wherein the glimpse captured by reading from the input image is an image patch generated by applying an array of Gaussian filters to the image.

9. The neural network system of claim 8 , wherein the parameters for applying the array of Gaussian filters are generated by applying a linear transformation to the decoder hidden state vector for the preceding time step.

10. The neural network system of claim 1 , wherein generating the decoder input for the decoder neural network from the encoder hidden state vector at the time step comprises:
using the encoder hidden state vector for the time step to parameterize a distribution of a latent vector; and
sampling the decoder input from the distribution.

11. The neural network system of claim 1 , wherein the subsystem is further configured to cause the decoder neural network to process a sequence of inputs sampled from a prior distribution to generate a new output image.

12. A computer implemented method for processing an input image through a recurrent encoder neural network and a recurrent decoder neural network to generate a final output image by repeatedly updating a neural network output at each of a plurality of time steps based on an input image to generate a final neural network output and generating the final output image from the final neural network output, the updating comprising, for each of the time step:
reading a glimpse from the input image using a decoder hidden state vector of the decoder neural network for the preceding time step;
providing, as input to the encoder neural network, the (i) glimpse and (ii) decoder hidden state vector for the preceding time step for processing;
receiving, as output from the encoder neural network, a generated encoder hidden state vector for the time step;
generating a decoder input for the decoder neural network from the generated encoder hidden state vector at the time step;
providing, as input to the decoder neural network, the decoder input for processing;
receiving, as output from the decoder neural network, a generated a decoder hidden state vector for the time step;
generating a neural network output update for the time step from the decoder hidden state vector for the time step; and
combining the neural network output update for the time step with a current neural network output to generate an updated neural network output.

13. The method of claim 12 , further comprising providing the encoder hidden state vectors from each of the time steps for a particular image as features of the particular image.

14. The method of claim 12 , wherein generating the decoder input for the decoder neural network from the encoder hidden state vector at the time step comprises:
using the encoder hidden state vector for the time step to parameterize a distribution of a latent vector; and
sampling the decoder input from the distribution.

15. The method of claim 12 , further comprising causing the decoder neural network to process a sequence of inputs sampled from a prior distribution to generate a new output image.

16. A computer storage medium encoded with instructions that, when executed by one or more computers, cause one or more computers to implement a neural network system, the neural network system comprising:
an encoder neural network, wherein the encoder neural network is a recurrent neural network that is configured to, for each input image processed by the encoder neural network and at each time step of a plurality steps:
receive a glimpse captured by reading from the input image,
receive a decoder hidden state vector of a decoder neural network for the preceding time step, and
process the glimpse, the decoder hidden state vector, and an encoder hidden state vector of the encoder neural network from the preceding time step to generate an encoder hidden state vector for the time step;

a decoder neural network, wherein the decoder neural network is a recurrent neural network that is configured to, for each of the plurality of time steps:
receive a decoder input for the time step, and
process the decoder hidden state vector for the preceding time step and the decoder input to generate a decoder hidden state vector for the time step; and

a subsystem, wherein the subsystem is configured to generate a final outp

Assignees

• Deepmind Tech Ltd

Inventors

• Gregor Karol
• Danihelka Ivo

Classifications

• G06N3/044 Primary

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/0455 Primary

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title