Determining optimal augmentations for a training data set

Patent metadata

Field	Value
Publication number	US-2021287084-A1
Application number	US-202016818891-A
Country	US
Kind code	A1
Filing date	Mar 13, 2020
Priority date	Mar 13, 2020
Publication date	Sep 16, 2021
Grant date	—

Abstract

A computer-implemented method according to one embodiment includes receiving a training data set to be applied to a model; selecting a subset of the training data set as a sample set; for each of a plurality of predetermined augmentations, applying the predetermined augmentation to the sample set to create an augmented sample set, training the model with the augmented sample set, determining a performance of the trained model, and assigning a weight to the predetermined augmentation for the training data set, based on the determined performance; and selecting one or more of the plurality of predetermined augmentations to be applied to the training data set before the training data set is applied to the model, based on the weight assigned to each of the plurality of predetermined augmentations.

First claim

What is claimed is:

1 . A computer-implemented method, comprising:
receiving a training data set to be applied to a model; selecting a subset of the training data set as a sample set; for each of a plurality of predetermined augmentations:
applying the predetermined augmentation to the sample set to create an augmented sample set,
training the model with the augmented sample set,
determining a performance of the trained model, and
assigning a weight to the predetermined augmentation for the training data set, based on the determined performance; and
selecting one or more of the plurality of predetermined augmentations to be applied to the training data set before the training data set is applied to the model, based on the weight assigned to each of the plurality of predetermined augmentations.

2 . The computer-implemented method of claim 1 , further comprising:
receiving a current training data set to be applied to the model; identifying characteristics of the current training data set; comparing the characteristics of the current training data set to characteristics of each of a plurality of preexisting training data sets; determining one of the plurality of preexisting training data sets having a highest similarity to the current training data set, based on the comparing; identifying weights assigned to each of a plurality of predetermined augmentations for the determined preexisting training data set having the highest similarity to the training data set; and selecting one or more of the plurality of predetermined augmentations for the determined preexisting training data set to be applied to the current training data set before the current training data set is applied to the model, based on the weights.

3 . The computer-implemented method of claim 1 , wherein the model includes a neural network.

4 . The computer-implemented method of claim 1 , wherein selecting the subset of the training data set includes identifying a predetermined percentage of the training data set as the sample set.

5 . The computer-implemented method of claim 1 , wherein each of the plurality of predetermined augmentations includes a process that changes one or more aspects of each data instance within the training data set to create an augmented data set.

6 . The computer-implemented method of claim 1 , wherein for each of the predetermined augmentations, the augmented sample set is input to the model as training data.

7 . The computer-implemented method of claim 1 , wherein for each of the predetermined augmentations, the weight assigned to the predetermined augmentation is determined based on a comparison of a first output of the model trained with the augmented sample set to a second output of another instance of the model trained with a non-augmented sample set.

8 . The computer-implemented method of claim 1 , wherein, for each of a plurality of predetermined augmentations, the weight for the predetermined augmentation is assigned to a grouping of the predetermined augmentation, the training data set, and the model.

9 . The computer-implemented method of claim 1 , wherein the plurality of predetermined augmentations are ranked based on their associated weights, and a predetermined number of augmentations having the highest weights are selected to be applied to the training data set to create an augmented training data set.

10 . The computer-implemented method of claim 9 , wherein the augmented training data set may be input into the model to train the model to perform one or more operations.

11 . The computer-implemented method of claim 1 , further comprising implementing a deep reinforcement method that leverages validation loss to assign positive and negative rewards to the plurality of predetermined augmentations to learn optimal augmentation configurations for the model.

12 . The computer-implemented method of claim 1 , wherein the plurality of predetermined augmentations is applied to the sample set by automatically forking a large varied set of virtual training instances in a cloud environment to validate which of the plurality of predetermined augmentations has a highest performance for the sample set.

13 . A computer program product for determining optimal augmentations for a training data set, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, wherein the computer readable storage medium is not a transitory signal per se, the program instructions executable by a processor to cause the processor to perform a method comprising:
receiving, by the processor, a training data set to be applied to a model; selecting, by the processor, a subset of the training data set as a sample set; for each of a plurality of predetermined augmentations:
applying, by the processor, the predetermined augmentation to the sample set to create an augmented sample set,
training, by the processor, the model with the augmented sample set,
determining, by the processor, a performance of the trained model, and
assigning, by the processor, a weight to the predetermined augmentation for the training data set, based on the determined performance; and
selecting, by the processor, one or more of the plurality of predetermined augmentations to be applied to the training data set before the training data set is applied to the model, based on the weight assigned to each of the plurality of predetermined augmentations.

14 . The computer program product of claim 13 , further comprising:
receiving, by the processor, a current training data set to be applied to the model; identifying, by the processor, characteristics of the current training data set; comparing, by the processor, the characteristics of the current training data set to characteristics of each of a plurality of preexisting training data sets; determining, by the processor, one of the plurality of preexisting training data sets having a highest similarity to the current training data set, based on the comparing; identifying, by the processor, weights assigned to each of a plurality of predetermined augmentations for the determined preexisting training data set having the highest similarity to the training data set; and selecting, by the processor, one or more of the plurality of predetermined augmentations for the determined preexisting training data set to be applied to the current training data set before the current training data set is applied to the model, based on the weights.

15 . The computer program product of claim 13 , wherein the model includes a neural network.

16 . The computer program product of claim 13 , wherein selecting the subset of the training data set includes identifying a predetermined percentage of the training data set as the sample set.

17 . The computer program product of claim 13 , wherein each of the plurality of predetermined augmentations includes a process that changes one or more aspects of each data instance within the training data set to create an augmented data set.

18 . The computer program product of claim 13 , wherein for each of the predetermined augmentations, the augmented sample set is input to the model as training data.

19 . The computer program product of claim 13 , wherein for each of the predetermined augmentations, the weight assigned to the predetermined augmentation is determined based on a comparison of a first output of the model trained with the augmented sample set to a second output of another instance of the model trained with

Assignees

• IBM

Inventors

• Sivakumar Gandhi
• Ekambaram Vijay
• Sivaswamy Hemant Kumar

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/092

  Reinforcement learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/02

  Neural networks · CPC title