Cooperative execution of a genetic algorithm with an efficient training algorithm for data-driven model creation

Patent metadata

Field	Value
Publication number	US-11100403-B2
Application number	US-201715663488-A
Country	US
Kind code	B2
Filing date	Jul 28, 2017
Priority date	Apr 17, 2017
Publication date	Aug 24, 2021
Grant date	Aug 24, 2021

Abstract

A method includes determining a trainable model to provide to a trainer, the trainable model determined based on modification of one or more models of a plurality of models. The plurality of models is generated based on a genetic algorithm and corresponds to a first epoch of the genetic algorithm. Each of the plurality of models includes data representative of a neural network. The method also includes providing the trainable model to the trainer. The method further includes adding a trained model, output by the trainer based on the trainable model, as input to a second epoch of the genetic algorithm, the second epoch subsequent to the first epoch.

First claim

What is claimed is:

1. A computer system comprising:
a memory to store an input data set and a plurality of data structures, each of the plurality of data structures including data representative of a neural network; and
a processor to execute a recursive search, wherein executing the recursive search comprises, during a first iteration:
determining a trainable data structure based on modification of one or more data structures of the plurality of data structures; and
providing the trainable data structure to an optimization trainer to:
train the trainable data structure based on a portion of the input data set to generate a trained data structure, wherein the optimization trainer, when executed, updates connection weights of the trainable data structure without updating a topology or activation functions of the trainable data structure; and
provide the trained data structure as input to a second iteration of the recursive search, the second iteration subsequent to the first iteration.

2. The computer system of claim 1 , wherein executing the recursive search further comprises, during the first iteration, selecting the one or more data structures based on their respective fitness values.

3. The computer system of claim 1 , wherein the modification corresponds to at least one of a crossover operation or a mutation operation with respect to the one or more data structures.

4. The computer system of claim 1 , wherein the optimization trainer is executed on a different device, graphics processing unit (GPU), processor, core, thread, or any combination thereof, than the recursive search.

5. A method comprising:
determining, by a processor of a computing device, a trainable model to provide to an optimization trainer, the trainable model determined based on modification of one or more models of a plurality of models, the plurality of models generated based on a genetic algorithm and corresponding to a first epoch of the genetic algorithm, wherein each of the plurality of models includes data representative of a neural network;
providing the trainable model to the optimization trainer to update connection weights of the trainable model without updating a topology or activation functions of the trainable model; and
adding a trained model, output by the optimization trainer based on the trainable model, as input to a second epoch of the genetic algorithm, the second epoch subsequent to the first epoch.

6. The method of claim 5 , further comprising selecting the one or more models based on their respective fitness values.

7. The method of claim 5 , wherein the trainable model is generated by performing at least one of a crossover operation or a mutation operation with respect to the one or more models.

8. The method of claim 5 , wherein the optimization trainer uses a portion of an input data set associated with the genetic algorithm to train the trainable model.

9. The method of claim 5 , wherein the data representative of the neural network includes node data corresponding to a plurality of nodes of the neural network and connection data corresponding to one or more connections of the neural network, wherein the node data corresponding to a particular node includes an activation function, an aggregation function, a bias, or any combination thereof, and wherein the connection data corresponding to a particular connection includes node pairs, the connection weights, or any combination thereof.

10. The method of claim 5 , wherein the first epoch is an initial epoch of the genetic algorithm.

11. The method of claim 5 , wherein the first epoch is a non-initial epoch of the genetic algorithm.

12. The method of claim 5 , wherein the second epoch and the first epoch are separated by at least one intervening epoch.

13. The method of claim 5 , wherein each of the plurality of models includes at least one output node to generate an output value corresponding to a field of an input data set associated with the genetic algorithm, and wherein a fitness value of a particular model is based at least partially on a comparison of the output value and the field of the input data set.

14. The method of claim 5 , wherein each of the plurality of models includes at least one output node to generate a classifier result.

15. The method of claim 5 , wherein the plurality of models corresponds to an input population of the first epoch, and further comprising:
grouping the models of the plurality of models into species based on genetic distance between the models;
determining a species fitness value of each of the species;
selectively removing one or more species from the genetic algorithm responsive to determining that the one or more species satisfy a stagnation criterion;
determining one or more elite species based on their respective species fitness values;
identifying elite members of each elite species; and
generating an output population to be input into the second epoch, wherein the output population includes each of the elite members, and at least one of a first model generated based on intra-species reproduction or a second model generated based on inter-species reproduction.

16. The method of claim 15 , wherein the output population further includes the trained model output by the optimization trainer.

17. A non-transitory computer-readable storage device storing instructions that, when executed, cause a computer to perform operations comprising:
determining a trainable model to provide to a trainer, the trainable model determined based on modification of one or more models of a plurality of models, the plurality of models generated based on a genetic algorithm and corresponding to a first epoch of the genetic algorithm, wherein each of the plurality of models includes data representative of a neural network;
providing the trainable model to the trainer, wherein the trainer updates connection weights of the trainable model without updating a topology or activation functions of the trainable model; and
adding a trained model, output by the trainer based on the trainable model, as input to a second epoch of the genetic algorithm, the second epoch subsequent to the first epoch.

18. The non-transitory computer-readable storage device of claim 17 , wherein the trainer comprises a backpropagation trainer.

19. The non-transitory computer-readable storage device of claim 17 , wherein:
the operations further comprise selecting the one or more models based on their respective fitness values;
the trainable model is generated by performing at least one of a crossover operation or a mutation operation with respect to the one or more models; or
a combination thereof.

20. A computer system comprising:
a memory to store an input data set and a plurality of data structures, each of the plurality of data structures including data representative of a neural network; and
a processor to execute a recursive search, wherein executing the recursive search comprises, during a first iteration:
determining a trainable data structure based on modification of one or more data structures of the plurality of data structures, wherein each of the plurality of data structures includes at least one output node to generate an output value corresponding to a field of the input data set associated with the recursive search, and wherein a fitness value of a particular data structure is based at least partially on a comparison of the output value and the field of the input data set; and
providing the trainable data structure to an optimization trainer to:

Assignees

• Sparkcognition Inc

Inventors

• Andoni Sari
• Moore Keith D
• Husain Syed Mohammad Amir

Classifications

• G06N3/0495

  Quantised networks; Sparse networks; Compressed networks · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06N3/0985

  Hyperparameter optimisation; Meta-learning; Learning-to-learn · CPC title

• G06N3/098

  Distributed learning, e.g. federated learning · CPC title

• G06N3/09

  Supervised learning · CPC title