Predicting likelihoods of conditions being satisfied using recurrent neural networks

• US 9,646,244 B2
• Filed: 05/09/2016
• Issued: 05/09/2017
• Est. Priority Date: 07/27/2015
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Corrado, Gregory Sean, Sutskever, Ilya, Dean, Jeffrey Adgate
• Primary Examiner(s): Starks, Wilbert L
• Application Number:
  US15/150,091
• Publication Number:
  US 20170032242A1
• Time in Patent Office: 365 Days
• Field of Search: None
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G06N 3/042 Knowledge-based neural networks; Logical representations of neural networks
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for predicting likelihoods of conditions being satisfied using recurrent neural networks. One of the systems is configured to process a temporal sequence comprising a respective input at each of a plurality of time steps and comprises: one or more recurrent neural network layers; one or more logistic regression nodes, wherein each of the logistic regression nodes corresponds to a respective condition from a predetermined set of conditions, and wherein each of the logistic regression nodes is configured to, for each of the plurality of time steps: receive the network internal state for the time step; and process the network internal state for the time step in accordance with current values of a set of parameters of the logistic regression node to generate a future condition score for the corresponding condition for the time step.

Claims

1. A recurrent neural network system configured to process a temporal sequence associated with a patient, the recurrent neural network system comprising data identifying a respective health event at each of a plurality of time steps, and the recurrent neural network system comprising:

one or more recurrent neural network layers, wherein the one or more recurrent neural network layers are configured to, for each of the plurality of time steps;

receive the data identifying the health event at the time step; and

collectively process the data identifying the health event at the time step to generate a network internal state for the time step;

one or more logistic regression nodes, wherein each of the logistic regression nodes corresponds to a respective condition from a predetermined set of conditions, and wherein each of the logistic regression nodes is configured to, for each of the plurality of time steps;

receive the network internal state for the time step; and

process the network internal state for the time step in accordance with current values of a set of parameters of the logistic regression node to generate a future condition score for the corresponding condition for the time step, wherein the future condition score for the corresponding condition represents a likelihood that the corresponding condition will be satisfied within a specified time period of the health event identified at the time step occurring.

2. The recurrent neural network system of claim 1, further comprising:

an output layer, wherein the output layer is configured to, for each of the plurality of time steps;

receive the network internal state for the time step; and

process the network internal state for the time step in accordance with current values of a set of parameters of the output layer to generate a respective next input score for each of a plurality of possible next health events, wherein the respective next input score for each of the plurality of possible next health events represents a likelihood that the possible next health event will be identified at a next time step in the temporal sequence.

3. The recurrent neural network system of claim 1, wherein each of the one or more recurrent neural network layers is configured to, for each of the plurality of time steps:

receive a layer input for the time step; and

process the layer input for the time step and a layer internal state for the recurrent neural network layer for a preceding time step in accordance with a current values of a set of parameters for the recurrent neural network layer to generate a layer internal state for the recurrent neural network layer for the time step.

4. The recurrent neural network system of claim 3, wherein the recurrent neural network system includes only one recurrent neural network layer, and wherein the layer internal state for the recurrent neural network layer for the time step is the network internal state for the time step.

5. The recurrent neural network system of claim 3, wherein the recurrent neural network system includes a plurality of recurrent neural network layers arranged in a sequence, and wherein the network internal state for the time step is a layer internal state of a highest recurrent neural network layer in the sequence for the time step.

6. The recurrent neural network system of claim 3, wherein the recurrent neural network system includes a plurality of recurrent neural network layers, and wherein the network internal state for the time step is combination of the layer internal states of the recurrent neural network layers for the time step.

7. The recurrent neural network system of claim 1, wherein each health event identified in the temporal sequence is selected from a set of possible health events.

8. The recurrent neural network system of claim 7, wherein the predetermined set of conditions includes conditions that are satisfied when events that are not in the set of possible health events occur within the specified time period of the health event identified at the time step.

9. The recurrent neural network system of claim 8, wherein the predetermined set of conditions also includes conditions that are satisfied when events that are in the set of possible health events occur within the specified time period of the input health event identified at the time step.

10. A method for processing a temporal sequence associated with a patient using a recurrent neural network system comprising one or more recurrent neural network layers and one or more logistic regression nodes, the temporal sequence comprising data identifying a respective health event at each of a plurality of time steps, and the method comprising, for each of the plurality of time steps:

receiving the data identifying the health event at the time step;

processing the data identifying the health event at the time step through the one or more recurrent neural network layers to generate a network internal state for the time step; and

processing the network internal state using each of the one or more logistic regression nodes, wherein each of the logistic regression nodes corresponds to a respective condition from a predetermined set of conditions, and wherein processing the network internal state using each of the one or more logistic regression nodes comprises;

processing the network internal state for the time step using the logistic regression node in accordance with current values of a set of parameters of the logistic regression node to generate a future condition score for the corresponding condition for the time step, wherein the future condition score for the corresponding condition represents a likelihood that the corresponding condition will be satisfied within a specified time period of the data identifying the health event at the time step occurring.

11. The method of claim 10, wherein the recurrent neural network system further comprises an output layer, and wherein the method further comprises, for each of the plurality of time steps:

processing the network internal state for the time step using the output layer in accordance with current values of a set of parameters of the output layer to generate a respective next input score for each of a plurality of possible next health events, wherein the respective next input score for each of the plurality of possible next health events represents a likelihood that the possible next health event will be identified at a next time step in the temporal sequence.

12. The method of claim 10, wherein each of the one or more recurrent neural network layers is configured to, for each of the plurality of time steps:

receive a layer input for the time step; and

process the layer input for the time step and a layer internal state for the recurrent neural network layer for a preceding time step in accordance with a current values of a set of parameters for the recurrent neural network layer to generate a layer internal state for the recurrent neural network layer for the time step.

13. The method of claim 12, wherein the recurrent neural network system includes only one recurrent neural network layer, and wherein the layer internal state for the recurrent neural network layer for the time step is the network internal state for the time step.

14. The method of claim 12, wherein the recurrent neural network system includes a plurality of recurrent neural network layers arranged in a sequence, and wherein the network internal state for the time step is a layer internal state of a highest recurrent neural network layer in the sequence for the time step.

15. The method of claim 12, wherein the recurrent neural network system includes a plurality of recurrent neural network layers, and wherein the network internal state for the time step is combination of the layer internal states of the recurrent neural network layers for the time step.

16. The method of claim 10, wherein each health event identified in the temporal sequence is selected from a set of possible health events.

17. The method of claim 16, wherein the predetermined set of conditions includes conditions that are satisfied when events that are not in the set of possible health events occur within a specified time period of the health event identified at the time step.

18. The method of claim 10, wherein the predetermined set of conditions also includes conditions that are satisfied when events that are in the set of possible health events occur within the specified time period of the input health event identified at the time step.

19. A computer program product encoded on one or more non-transitory computer readable media, the computer program product comprising instructions that when executed by one or more computers cause the one or more computers to perform operations to implement a recurrent neural network system configured to process a temporal sequence associated with a patient, the recurrent neural network system comprising data identifying a respective health event at each of a plurality of time steps, and the recurrent neural network system comprising:

one or more recurrent neural network layers, wherein the one or more recurrent neural network layers are configured to, for each of the plurality of time steps;

receive the data identifying the health event at the time step; and

collectively process the data identifying the health event at the time step to generate a network internal state for the time step;

one or more logistic regression nodes, wherein each of the logistic regression nodes corresponds to a respective condition from a predetermined set of conditions, and wherein each of the logistic regression nodes is configured to, for each of the plurality of time steps;

receive the network internal state for the time step; and

process the network internal state for the time step in accordance with current values of a set of parameters of the logistic regression node to generate a future condition score for the corresponding condition for the time step, wherein the future condition score for the corresponding condition represents a likelihood that the corresponding condition will be satisfied within a specified time period of the health event identified at the time step occurring.

20. The computer program product of claim 19, wherein the recurrent neural network system further comprises an output layer, and wherein the operations further comprise, for each of the plurality of time steps:

processing the network internal state for the time step using the output layer in accordance with current values of a set of parameters of the output layer to generate a respective next input score for each of a plurality of possible next health events, wherein the respective next input score for each of the plurality of possible next health events represents a likelihood that the possible next health event will be identified at a next time step in the temporal sequence.