Dynamic Data Augmentation Based on Sample Hardness

Solution Overview

Problem

Existing data augmentation techniques often result in over-augmentation, leading to unnatural data distributions and compromised learnable features, as they apply a single augmentation scheme to all data samples without considering their individual hardness in classification.

Innovation Solution

A dynamic augmentation approach that adjusts augmentation strength based on the hardness of each data sample, applying increased augmentation to 'easy' samples and decreased augmentation to 'hard' samples to maintain discriminative features and improve model performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If a single augmentation scheme is applied to all data samples, then the implementation is simple and uniform, but it causes over-augmentation for easy samples and insufficient augmentation for hard samples, leading to loss of learnable features

Engineering Contradiction:

Improvesimplicity of augmentation implementationVSAvoidloss of learnable features

Core Design Contradiction:

Ease of manufactureVSLoss of information

Solution Approach 1:

The patent applies different augmentation strengths to different data samples based on their individual hardness characteristics. Easy samples receive stronger augmentation while hard samples receive weaker augmentation, making the augmentation strategy locally adapted rather than uniformly applied. This resolves the contradiction by maintaining simplicity through automated hardness-based differentiation while preventing feature loss in hard samples.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent dynamically adjusts augmentation strength based on the measured hardness of each data sample. The augmentation parameter is not fixed but varies dynamically according to sample-specific characteristics, allowing the system to adapt augmentation intensity to match the actual learning needs of each sample while maintaining implementation simplicity through automated adjustment.

Inventive Principle:

Principle #15Dynamics

2Reliability

If strong augmentation is applied to all samples to improve model generalization, then model performance may improve, but hard samples lose their discriminative features and learnable signals

Engineering Contradiction:

Improvemodel generalization performanceVSAvoidcompromised learnable features in hard samples

Core Design Contradiction:

ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The patent applies different augmentation strengths to different data samples based on their individual hardness characteristics. Easy samples receive stronger augmentation while hard samples receive weaker augmentation, making the augmentation strategy locally adapted rather than uniformly applied. This resolves the contradiction by maintaining simplicity through automated hardness-based differentiation while preventing feature loss in hard samples.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent changes the augmentation parameter (strength) based on the hardness characteristic of each sample. By adjusting this key parameter dynamically, the system achieves strong augmentation for samples that can tolerate it while protecting hard samples from excessive augmentation, thus improving generalization without compromising learnable features.

Inventive Principle:

Principle #35Parameter changes

3Loss of information

If weak augmentation is applied to all samples to preserve features, then learnable features are maintained, but easy samples are not sufficiently challenged to improve model performance

Engineering Contradiction:

Improvepreservation of learnable featuresVSAvoidmodel generalization performance

Core Design Contradiction:

Loss of informationVSReliability

Solution Approach 1:

The patent applies different augmentation strengths to different data samples based on their individual hardness characteristics. Easy samples receive stronger augmentation while hard samples receive weaker augmentation, making the augmentation strategy locally adapted rather than uniformly applied. This resolves the contradiction by maintaining simplicity through automated hardness-based differentiation while preventing feature loss in hard samples.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent dynamically adjusts augmentation strength based on the measured hardness of each data sample. The augmentation parameter is not fixed but varies dynamically according to sample-specific characteristics, allowing the system to apply strong augmentation to easy samples for better challenging while protecting hard samples, thus achieving both feature preservation and improved generalization.

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS20220036212A1Dynamic augmentation based on data sample hardness

Publication Date: 2022.02.03 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

In an approach for dynamic augmentation based on data sample hardness for training a learning model, a processor defines one or more augmentations for a dataset for training the learning model. A processor applies the one or more augmentations to the dataset. A processor trains the learning model with the one or more augmentations. A processor measures hardness of one or more data samples in the dataset. A processor adjusts the one or more augmentations for the one or more data samples based on corresponding hardness of the one or more data samples. A processor applies the adjusted one or more augmentations to the dataset. A processor trains the learning model with the adjusted one or more augmentations applied to the dataset.