Dataset Weighting via Feature Space Proximity for ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Datasets for training machine learning models often contain identical or very similar data points and unbalanced distributions within classes, leading to limitations in learning capability and model performance.
Innovation Solution
A method for weighting a dataset by determining the proximity of each data point to surrounding points in a feature space and assigning weighting values based on this proximity, which are then used during model training to balance the dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data points are removed or classes are balanced by different weighting in prior art, then dataset imbalance is reduced, but learning capability is limited due to unequal distribution within classes
Solution Approach 1:
The patent applies local quality by assigning different weighting values to individual data points based on their local density in the feature space. Instead of uniform class-level weighting, each data point receives a customized weight reflecting its proximity to neighbors, thereby addressing local distribution imbalances while preserving global dataset structure and learning capability.
Solution Approach 2:
The patent changes the parameter of data point weighting from uniform class-level weights to variable instance-level weights determined by local density measurements. This parameter transformation enables the model to adapt to local distribution variations within classes, improving both reliability and learning capability simultaneously.
2Reliability
If further data is collected or model architecture is modified to balance the dataset, then dataset balance improves, but computational complexity and training cost increase
Solution Approach 1:
The patent employs self-service by using the model's own feature space and local density measurements to automatically determine weighting values for training data. This self-generated weighting mechanism eliminates the need for external data collection or complex architectural modifications, achieving dataset balance through an efficient, model-intrinsic process.
Solution Approach 2:
The patent substitutes mechanical approaches (data collection, architecture modification) with a computational weighting mechanism based on local density analysis. This substitution replaces resource-intensive physical operations with efficient mathematical calculations, reducing computational complexity while maintaining dataset balance.
3Productivity
If identical data points are removed to reduce redundancy, then dataset efficiency improves, but data distribution uniformity worsens
Solution Approach 1:
The patent applies partial action by retaining all data points but applying partial weighting based on local density. Instead of completely removing redundant points, it applies a partial weighting factor that reflects their similarity to neighbors, thereby maintaining data distribution uniformity while still improving training efficiency through selective emphasis on diverse samples.
Data Source
AI summary
The invention relates to a method for weighting a dataset for training a machine learning model, comprising the following steps:ascertaining (101) a position of each data point (1) of the dataset in a feature space (2) of a further machine learning model,ascertaining (102) a respective proximity of the data points (1) to at least one surrounding data point (1′) in the feature space (2) of the further machine learning model based on the position of the data points (1) ascertained,determining (103) a weighting value for each data point (1) based on the proximity to the at least one surrounding data point (1′) ascertained,using (104) the determined weighting values of the data points (1) when training the machine learning model, wherein the weighting values determined are used for an extent of consideration of the respective data points (1) in the training.The invention further relates to a computer program, a device, and a storage medium for this purpose.