Training machine learning models by determining update rules using neural networks

1. A computer-implemented method for training a target machine learning model having a plurality of model parameters using an optimizer neural network having a plurality of optimizer parameters, the method comprising, at each current iteration of a plurality of iterations:

determining a respective update rule for each of the plurality of model parameters using the optimizer neural network by operating the optimizer neural network independently on each of the plurality of model parameters of the target machine learning model, the determining comprising, for each respective model parameter:

generating a respective parameter-specific input that comprises a gradient of a target objective function of the target machine learning model with respect to the respective model parameter; and

processing the respective parameter-specific input using the optimizer neural network and in accordance with current values of the optimizer parameters to generate a respective optimizer output that specifies a respective parameter-specific update rule for updating the respective model parameter;

applying the update rules generated by the optimizer neural network to the model parameters of the target machine learning model to update values of the model parameters; and

updating the current values of the optimizer parameters by using gradient descent techniques to minimize an optimizer objective function that depends at least on a function value of the target objective function of the target machine learning model computed using the values of the model parameters of the target machine learning model that have been updated at the current iteration.

2. The method of claim 1 , wherein applying the determined update rule for a final iteration in the plurality of iterations to the model parameters generates trained model parameters.

3. The method of claim 1 , wherein the target machine learning model comprises a neural network.

4. The method of claim 1 , wherein the current iteration is an iteration after the first iteration of the plurality of iterations, and the optimizer objective function further depends on the values of the model parameters at one or more iterations that precede the current iteration.

5. The method of claim 1 , wherein the optimizer neural network is a recurrent neural network (RNN).

6. The method of claim 5 , further comprising, at each of the plurality of iterations, providing a previous hidden state of the RNN as input to the RNN for the iteration.

7. The method of claim 5 , wherein the optimizer neural network is a long short-term memory (LSTM) neural network.

8. The method of claim 1 , further comprising preprocessing the respective parameter-specific inputs to the optimizer neural network to disregard gradients that are smaller than a predetermined threshold.

9. The method of claim 1 , wherein the optimizer objective function further depends on function values computed by applying the target objective function of the target machine learning model to the values of the model parameters that have been computed at one or more iterations that precede the current iteration.

10. The method of claim 9 , wherein the optimizer objective function is computed at the current iteration based on a weighted sum of function values of the target objective function of the target machine learning model, wherein each function value is computed by applying the target objective function to the values of the model parameters that have been computed at the current iteration or at an iteration that precede the current iteration.

11. A system comprising one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations for training a target machine learning model having a plurality of model parameters using an optimizer neural network having a plurality of optimizer parameters, the operations comprising, at each current iteration of a plurality of iterations:

determining a respective update rule for each of the plurality of model parameters using the optimizer neural network by operating the optimizer neural network independently on each of the plurality of model parameters of the target machine learning model, the determining comprising, for each respective model parameter:

generating a respective parameter-specific input that comprises a gradient of a target objective function of the target machine learning model with respect to the respective model parameter; and

processing the respective parameter-specific input using the optimizer neural network and in accordance with current values of the optimizer parameters to generate a respective optimizer output that specifies a respective parameter-specific update rule for updating the respective model parameter;

applying the update rules generated by the optimizer neural network to the model parameters of the target machine learning model to update values of the model parameters; and

updating the current values of the optimizer parameters by using gradient descent techniques to minimize an optimizer objective function that depends at least on a function value of the target objective function of the target machine learning model computed using the values of the model parameters that have been updated at the current iteration.

12. The system of claim 11 , wherein applying the determined update rule for a final iteration in the plurality of iterations to the model parameters generates trained model parameters.

13. The system of claim 11 , wherein the target machine learning model comprises a neural network.

14. The system of claim 11 , wherein the optimizer objective function further depends on the values of the model parameters at one or more iterations that precede the current iteration.

15. The system of claim 11 , wherein the optimizer neural network is a recurrent neural network (RNN).

16. One or more non-transitory computer-readable storage media storing instructions that, when executed by one or more computers, cause the one or more computers to perform operations for training a target machine learning model having a plurality of model parameters using an optimizer neural network having a plurality of optimizer parameters, the operations comprising, at each current iteration of a plurality of iterations:

determining a respective update rule for each of the plurality of model parameters using the optimizer neural network by operating the optimizer neural network independently on each of the plurality of model parameters of the target machine learning model, the determining comprising, for each respective model parameter:

generating a respective parameter-specific input that comprises a gradient of a target objective function of the target machine learning model with respect to the respective model parameter; and

processing the respective parameter-specific input using the optimizer neural network and in accordance with current values of the optimizer parameters to generate a respective optimizer output that specifies a respective parameter-specific update rule for updating the respective model parameter;

applying the update rules generated by the optimizer neural network to the model parameters of the target machine learning model to update values of the model parameters; and

updating the current values of the optimizer parameters by using gradient descent techniques to minimize an optimizer objective function that depends at least on a function value of the target objective function of the target machine learning model computed using the values of the model parameters that have been updated at the current iteration.

17. The one or more non-transitory computer-readable storage media of claim 16 , wherein applying the determined update rule for a final iteration in the plurality of iterations to the model parameters generates trained model parameters.

18. The one or more non-transitory computer-readable storage media of claim 16 , wherein the target machine learning model comprises a neural network.

19. The one or more non-transitory computer-readable storage media of claim 16 , wherein the optimizer objective function further depends on the values of the model parameters at one or more iterations that precede the current iteration.