Synthetic training data generation for improved machine learning model generalization capability

Abstract

The invention relates to synthetic training data generation for improving machine learning model generalization capability. Systems and techniques are provided that facilitate synthetic training data generation to achieve improved machine learning generalization capabilities. In various embodiments, an element enhancement component may generate a set of annotated preliminary training images based on an annotated source image. In various aspects, an annotated preliminary training image may be formed by inserting at least one element of interest or at least one background element into an annotated source image. In various cases, a modal enhancement component may generate a set of annotated intermediate training images based on the set of annotated preliminary training images. In various cases, an annotated intermediate training image may be formed by changing at least one modal-based characteristic of an annotated preliminary training image. In various aspects, a geometric enhancement component may generate a set of annotated deployable training images based on the set of annotated intermediate training images. In various cases, an annotated deployable training image may be formed by changing at least one geometric characteristic of an annotated intermediate training image. In various embodiments, a training component may train a machine learning model on the set of annotated deployable training images.

Description

technical field [0001] The subject disclosure relates generally to the training of machine learning models, and more specifically to the generation of synthetic training data for improving the generalization capabilities of machine learning models. Background technique [0002] The efficacy and / or generalization ability of a machine learning model depends on the realism, volume, variety and / or velocity of the data used to train the machine learning model. In other words, the specific implementation of high-quality, larger volume, more variation / diversity, and / or easier availability of training data can lead to the creation of machine learning models that are immune to the various challenges faced in real-world operational scenarios . Conversely, the implementation of low-quality, smaller-capacity, less-variable / poorly diverse, and / or less-available training data implementations can lead to the creation of machines that are susceptible to the various challenges faced in real...

AI Technical Summary

Problems solved by technology

Conversely, the implementation of low-quality, smaller-capacity, less-variable / poorly diverse, and / or less-available training data implementations can lead to the creation of machines that are susceptible to the various challenges faced in real-world operational scenarios learning model

Images