Dynamically updated neural network structures for content distribution networks

• US 9,336,483 B1
• Filed: 04/03/2015
• Issued: 05/10/2016
• Est. Priority Date: 04/03/2015
• Status: Active Grant

• Current Assignee: Pearson Education Incorporated (Pearson plc)
• Original Assignee: Pearson Education Incorporated (Pearson plc)
• Inventors: Abeysooriya, Thilani, Withana, Bhanuka, Liyanarachchi, Achila, Amaratunga, Thimira Dilina Kalindu
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Pellett, Daniel
• Application Number:
  US14/678,312
• Time in Patent Office: 403 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

Dynamically updating neural network systems may be implemented to generate, train, evaluate and update artificial neural network data structures used by content distribution networks. Such systems and methods described herein may include generating and training neural networks, using neural networks to perform predictive analysis and other decision-making processes within content distribution networks, evaluating the performance of neural networks, and generating and training pluralities of replacement candidate neural networks within cloud computing architectures and/or other computing environments.

Claims

1. A dynamically updating neural network system for evaluating and updating artificial neural networks for electronic learning systems, the dynamically updating neural network system comprising:

a database server comprising one or more databases that;

receive and store neural network training data corresponding to input and output data associated with an electronic learning system;

a network interface configured to provide one or more electronic learning system servers with access to the database server via one or more computer networks; and

a neural network management server of the electronic learning system comprising;

a processing unit comprising one or more processors; and

memory coupled with and readable by the processing unit and storing therein a set of instructions which, when executed by the processing unit, causes the neural network management server to;

retrieve, from the database server and via the network interface, first neural network training data corresponding to input and output data of the electronic learning system;

generate and train a first electronic learning system neural network using the first neural network training data;

determine an error threshold associated with the trained first electronic learning system neural network;

receive additional input data and corresponding output data associated with the electronic learning system;

execute a plurality of predictive analyses using the first electronic learning system neural network, based on the additional input data associated with the electronic learning system;

evaluate the first electronic learning system neural network by;

comparing the results of each of the plurality of predictive analyses with the corresponding additional output data;

aggregating the results of the plurality of predictive analyses to generate an aggregate error rate for the trained first electronic learning system neural network; and

determining whether the aggregate error rate has exceeded the error threshold associated with the trained first electronic learning system neural network;

in response to determining that the aggregate error rate has exceeded the error threshold associated with the trained first electronic learning system neural network, generate and train a plurality of additional electronic learning system neural networks using at least the first neural network training data and the additional input and output data associated with the electronic learning system;

evaluate the plurality of additional electronic learning system neural networks by executing one or more identical predictive analyses using each of the plurality of additional electronic learning system neural networks;

select a replacement electronic learning system neural network based on the evaluation of the plurality of additional electronic learning system neural networks; and

replace the first electronic learning system neural network with the selected replacement electronic learning system neural network.

2. The dynamically updating neural network system of claim 1, wherein generating and training the plurality of additional electronic learning system neural networks comprises allocating a set of computing resources in a cloud computing architecture to generate and train the plurality of additional electronic learning system neural networks.

3. The dynamically updating neural network system of claim 1, wherein the memory of the neural network management server further stores instructions which, when executed by the processing unit, causes the neural network management server to:

execute a genetic mutation algorithm to determine a structural mutation for at least one of the plurality of additional electronic learning system neural networks.

4. The dynamically updating neural network system of claim 1, wherein generating the plurality of additional electronic learning system neural networks comprises:

executing a randomization algorithm to assign an initial random weight value to each of a plurality of node interconnections within each of the additional electronic learning system neural networks.

5. The dynamically updating neural network system of claim 1, wherein the error threshold associated with the trained first electronic learning system neural network is determined based at least in part on a cost or an availability of computing resources in a cloud computing architecture.

6. The dynamically updating neural network system of claim 1, wherein the memory of the neural network management server further stores instructions which, when executed by the processing unit, causes the neural network management server to:

update the first neural network training data to include the additional input and output data of the electronic learning system, wherein the additional input and output data of the electronic learning system is received after the generation and training of the first electronic learning system neural network.

7. A method of evaluating and updating artificial neural networks for electronic learning systems, the method comprising:

retrieving, from a database server via a network interface, first neural network training data corresponding to input and output data of an electronic learning system;

generating and training a first electronic learning system neural network using the first neural network training data;

determining an error threshold associated with the trained first electronic learning system neural network;

receiving additional input data and corresponding output data associated with the electronic learning system;

executing a plurality of predictive analyses using the first electronic learning system neural network, based on the additional input associated with the electronic learning system;

evaluating the first electronic learning system neural network by;

comparing the results of each of the plurality of predictive analyses with the corresponding additional output data;

aggregating the results of the plurality of predictive analyses to generate an aggregate error rate for the trained first electronic learning system neural network; and

determining whether the aggregate error rate has exceeded the error threshold associated with the trained first electronic learning system neural network;

in response to determining that the aggregate error rate has exceeded the error threshold associated with the trained first electronic learning system neural network, generating and training a plurality of additional electronic learning system neural networks using at least the first neural network training data and the additional input and output data associated with the electronic learning system;

evaluating the plurality of additional electronic learning system neural networks by executing one or more identical predictive analyses using each of the plurality of additional electronic learning system neural networks;

selecting a replacement electronic learning system neural network based on the evaluation of the plurality of additional electronic learning system neural networks; and

replacing the first electronic learning system neural network with the selected replacement electronic learning system neural network.

8. The method of claim 7, wherein generating and training the plurality of additional electronic learning system neural networks comprises allocating a set of computing resources in a cloud computing architecture to generate and train the plurality of additional electronic learning system neural networks.

9. The method of claim 7, further comprising:

executing a genetic mutation algorithm to determine a structural mutation for at least one of the plurality of additional electronic learning system neural networks.

10. The method of claim 7, wherein generating the plurality of additional electronic learning system neural networks comprises:

executing a randomization algorithm to assign an initial random weight value to each of a plurality of node interconnections within each of the additional electronic learning system neural networks.

11. The method of claim 7, wherein the error threshold associated with the trained first electronic learning system neural network is determined based at least in part on a cost or an availability of computing resources in a cloud computing architecture.

12. The method of claim 7, further comprising:

updating the first neural network training data to include the additional input and output data associated with the electronic learning system, wherein the additional input and output data associated with the electronic learning system is received after the generation and training of the first electronic learning system neural network.

13. A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause one or more data processors to perform actions including:

retrieving, from a database server via a network interface, first neural network training data corresponding to input and output data of an electronic learning system;

generating and training a first electronic learning system neural network using the first neural network training data;

determining an error threshold associated with the trained first electronic learning system neural network;

receiving additional input data and corresponding output data associated with the electronic learning system;

executing a plurality of predictive analyses using the first electronic learning system neural network, based on the additional input associated with the electronic learning system;

evaluating the first electronic learning system neural network by;

comparing the results of each of the plurality of predictive analyses with the corresponding additional output data;

aggregating the results of the plurality of predictive analyses to generate an aggregate error rate for the trained first electronic learning system neural network; and

determining whether the aggregate error rate has exceeded the error threshold associated with the trained first electronic learning system neural network;

in response to determining that the aggregate error rate has exceeded the error threshold associated with the trained first electronic learning system neural network, generating and training a plurality of additional electronic learning system neural networks using at least the first neural network training data and the additional input and output data associated with the electronic learning system;

evaluating the plurality of additional electronic learning system neural networks by executing one or more identical predictive analyses using each of the plurality of additional electronic learning system neural networks;

selecting a replacement electronic learning system neural network based on the evaluation of the plurality of additional electronic learning system neural networks; and

replacing the first electronic learning system neural network with the selected replacement electronic learning system neural network.

14. The computer-program product as recited in claim 13, wherein generating and training the plurality of additional electronic learning system neural networks comprises allocating a set of computing resources in a cloud computing architecture to generate and train the plurality of additional electronic learning system neural networks.

15. The computer-program product as recited in claim 13, wherein the actions further include:

executing a genetic mutation algorithm to determine a structural mutation for at least one of the plurality of additional electronic learning system neural networks.

16. The computer-program product as recited in claim 13, wherein generating the plurality of additional electronic learning system neural networks comprises:

executing a randomization algorithm to assign an initial random weight value to each of a plurality of node interconnections within each of the additional electronic learning system neural networks.

17. The computer-program product as recited in claim 13, wherein the error threshold associated with the trained first electronic learning system neural network is determined based at least in part on a cost or an availability of computing resources in a cloud computing architecture.