Training neural networks using a prioritized experience memory

1. A method for training a neural network used to select actions performed by a reinforcement learning agent interacting with an environment by performing actions that cause the environment to transition states, the method comprising:

maintaining a replay memory, the replay memory storing pieces of experience data for use in training the neural network, wherein:

each piece of experience data has been generated as a result of the reinforcement learning agent interacting with the environment,

each piece of experience data comprises a respective current observation characterizing a respective current state of the environment, a respective current action performed by the agent in response to the current observation, a respective next observation characterizing a respective next state of the environment, and a reward received in response to the agent performing the current action,

a plurality of the pieces of experience data are each associated with a respective expected learning progress measure that (i) is a measure of an expected amount of progress that would be made in the training of the neural network if the neural network is trained on the piece of experience data and (ii) is derived from a result of a preceding time that the piece of experience data was used in training the neural network;

selecting a piece of experience data from the replay memory by prioritizing for selection pieces of experience data having relatively higher expected learning progress measures, comprising:

determining, based on the respective expected learning progress measures for the pieces of experience data, a respective probability for each of the pieces of experience data in the replay memory, and

sampling a piece of experience data from the replay memory in accordance with the determined probabilities;

training, using a reinforcement learning technique, the neural network on the selected piece of experience data; and

associating, in the replay memory, the selected piece of experience data with a new expected learning progress measure derived from a result of training the neural network on the selected piece of experience data.

2. The method of claim 1 , wherein determining, based on the respective expected learning progress measures for the pieces of experience data, a respective probability for each of the pieces of experience data in the replay memory comprises:

determining a respective probability for each piece of experience data such that pieces of experience data having higher expected learning progress measures have higher probabilities than pieces of experience data having relatively lower expected learning progress measures.

3. The method of claim 2 , wherein the probability P(i) for a piece of experience data i satisfies:

P

⁡

(

i

)

=

p

i

α

∑

k

⁢

p

k

α

where α is a predetermined constant, k ranges across the pieces of experience data in the replay memory, and p i is a priority for the piece of experience data i derived from the expected learning progress measure for the piece of experience data i.

4. The method of claim 3 , wherein the priority is the expected learning measure plus a constant value.

5. The method of claim 3 , wherein the priority is a fraction having a predetermined positive value as a numerator and a rank of the piece of experience data i in a ranking of the pieces of experience data in the replay memory according to their expected learning progress measures as a denominator.

6. The method of claim 3 , wherein the priority is set to a maximum value for a piece of experience data that has not yet been used in training in the neural network.

7. The method of claim 1 , wherein training the neural network on the selected piece of experience data comprises:

determining a temporal difference learning error for the selected piece of experience data; and

using the temporal difference learning error in adjusting values of the parameters of the neural network.

8. The method of claim 7 , wherein using the temporal difference learning error in adjusting the values of the parameters comprises:

determining a weight for the temporal difference learning error using the expected learning progress measure for the selected piece of experience data;

adjusting the temporal difference learning error using the weight; and

using the adjusted temporal difference learning error as a target error for adjusting the values of the parameters of the neural network.

9. The method of claim 8 , further comprising annealing an exponent used in computing the weight during the training of the neural network.

10. The method of claim 1 , wherein the expected learning progress measure for each piece of experience data in the replay memory is an absolute value of a temporal difference learning error determined for the piece of experience data the preceding time the experience tuple was used in training the neural network.

11. The method of claim 1 , wherein the expected learning progress measure for each piece of experience data in the replay memory is a derivative of an absolute value of a temporal difference learning error determined for the piece of experience data the preceding time the experience tuple was used in training the neural network.

12. The method of claim 1 , wherein the expected learning progress measure for each piece of experience data in the replay memory is a norm of an induced weight-change by using the piece of experience data to train the neural network.

13. The method of claim 1 , further comprising:

determining when to delete pieces of experience data from the replay memory using the expected learning progress measures.

14. The method of claim 13 , further comprising:

determining that (i) an expected progress measure for a particular piece of experience data or (ii) a value derived from the expected progress measure for the particular piece of experience data falls below a threshold; and

in response, deleting the particular piece of experience data from the replay memory.

15. The method of claim 1 , further comprising:

controlling the reinforcement learning agent using the neural network while the reinforcement learning agent interacts with the environment.

16. 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 a method for training a neural network used to select actions performed by a reinforcement learning agent interacting with an environment by performing actions that cause the environment to transition states, the method comprising:

maintaining a replay memory, the replay memory storing pieces of experience data for use in training the neural network, wherein:

each piece of experience data has been generated as a result of the reinforcement learning agent interacting with the environment,

each piece of experience data comprises a respective current observation characterizing a respective current state of the environment, a respective current action performed by the agent in response to the current observation, a respective next observation characterizing a respective next state of the environment, and a reward received in response to the agent performing the current action,

a plurality of the pieces of experience data are each associated with a respective expected learning progress measure that (i) is a measure of an expected amount of progress that would be made in the training of the neural network if the neural network is trained on the piece of experience data and (ii) is derived from a result of a preceding time that the piece of experience data was used in training the neural network;

selecting a piece of experience data from the replay memory by prioritizing for selection pieces of experience data having relatively higher expected learning progress measures, comprising:

determining, based on the respective expected learning progress measures for the pieces of experience data, a respective probability for each of the pieces of experience data in the replay memory, and

sampling a piece of experience data from the replay memory in accordance with the determined probabilities;

training, using a reinforcement learning technique, the neural network on the selected piece of experience data; and

associating, in the replay memory, the selected piece of experience data with a new expected learning progress measure derived from a result of training the neural network on the selected piece of experience data.

17. The system of claim 16 , determining, based on the respective expected learning progress measures for the pieces of experience data, a respective probability for each of the pieces of experience data in the replay memory comprises:

determining a respective probability for each piece of experience data such that pieces of experience data having higher expected learning progress measures have higher probabilities than pieces of experience data having relatively lower expected learning progress measures.

18. The system of claim 17 , wherein the probability P(i) for a piece of experience data i satisfies:

P

⁡

(

i

)

=

p

i

α

∑

k

⁢

p

k

α

where α is a predetermined constant, k ranges across the pieces of experience data in the replay memory, and p i is a priority for the piece of experience data i derived from the expected learning progress measure for the piece of experience data i.

19. The system of claim 17 , wherein training the neural network on the selected piece of experience data comprises:

determining a temporal difference learning error for the selected piece of experience data; and

using the temporal difference learning error in adjusting values of the parameters of the neural network.

20. A non-transitory computer storage medium encoded with instructions that, when executed by one or more computers, cause the one or more computers to perform operations for training a neural network used to select actions performed by a reinforcement learning agent interacting with an environment by performing actions that cause the environment to transition states, the operations comprising:

maintaining a replay memory, the replay memory storing pieces of experience data for use in training the neural network, wherein:

each piece of experience data has been generated as a result of the reinforcement learning agent interacting with the environment,

each piece of experience data comprises a respective current observation characterizing a respective current state of the environment, a respective current action performed by the agent in response to the current observation, a respective next observation characterizing a respective next state of the environment, and a reward received in response to the agent performing the current action,

a plurality of the pieces of experience data are each associated with a respective expected learning progress measure that (i) is a measure of an expected amount of progress that would be made in the training of the neural network if the neural network is trained on the piece of experience data and (ii) is derived from a result of a preceding time that the piece of experience data was used in training the neural network;

selecting a piece of experience data from the replay memory by prioritizing for selection pieces of experience data having relatively higher expected learning progress measures, comprising:

determining, based on the respective expected learning progress measures for the pieces of experience data, a respective probability for each of the pieces of experience data in the replay memory, and

sampling a piece of experience data from the replay memory in accordance with the determined probabilities;

training, using a reinforcement learning technique, the neural network on the selected piece of experience data; and

associating, in the replay memory, the selected piece of experience data with a new expected learning progress measure derived from a result of training the neural network on the selected piece of experience data.