Dynamic Sampling for Balanced ML Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for preparing training data sets for machine learning models often result in under-sampling of the highest resource category due to sampling without replacement, leading to an unbalanced target distribution and potential bias.
Innovation Solution
The use of indexes and counters to randomly sample a categorical data set, ensuring the highest resource category is fully sampled without replacement before sampling lower resource categories with replacement, thereby optimizing the target distribution for machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sampling without replacement is used to avoid duplicate data items, then artificial bias is reduced, but the highest resource category becomes under-sampled
Solution Approach 1:
The patent changes the sampling parameter by introducing a dynamic replacement policy that differs from traditional sampling methods. Instead of uniformly applying sampling without replacement, the system selectively applies replacement based on category resource levels, allowing the highest resource category to be sampled with replacement while lower resource categories are sampled without replacement. This parameter change resolves the contradiction by maintaining distribution balance while ensuring adequate sampling of the highest resource category.
Solution Approach 2:
The patent applies different sampling strategies to different categories based on their resource levels. The highest resource category receives special treatment with sampling with replacement, while lower resource categories use sampling without replacement. This local differentiation resolves the contradiction by tailoring the sampling approach to the specific needs of each category, ensuring that the highest resource category is adequately represented while still avoiding excessive duplication across the overall dataset.
2Quantity of substance
If sampling with replacement is used to maximize data item usage, then the highest resource category is fully sampled, but duplicate data items create artificial bias
Solution Approach 1:
The patent changes the sampling parameter by introducing a dynamic replacement policy that differs from traditional sampling methods. Instead of uniformly applying sampling with replacement, the system selectively applies replacement based on category resource levels, allowing the highest resource category to be sampled with replacement while lower resource categories are sampled without replacement. This parameter change resolves the contradiction by maintaining distribution balance while ensuring adequate sampling of the highest resource category.
Solution Approach 2:
The patent applies different sampling strategies to different categories based on their resource levels. The highest resource category receives special treatment with sampling with replacement, while lower resource categories use sampling without replacement. This local differentiation resolves the contradiction by tailoring the sampling approach to the specific needs of each category, ensuring that the highest resource category is adequately represented while still avoiding excessive duplication across the overall dataset.
3Ease of operation
If uniform sampling is applied to all categories, then sampling simplicity is maintained, but the highest resource category is under-sampled relative to lower resource categories
Solution Approach 1:
The patent applies different sampling strategies to different categories based on their resource levels. The highest resource category receives special treatment with sampling with replacement, while lower resource categories use sampling without replacement. This local differentiation resolves the contradiction by tailoring the sampling approach to the specific needs of each category, ensuring that the highest resource category is adequately represented while still avoiding excessive duplication across the overall dataset.
Solution Approach 2:
The patent changes the sampling parameter by introducing a dynamic replacement policy that differs from traditional sampling methods. Instead of uniformly applying sampling without replacement, the system selectively applies replacement based on category resource levels, allowing the highest resource category to be sampled with replacement while lower resource categories are sampled without replacement. This parameter change resolves the contradiction by maintaining distribution balance while ensuring adequate sampling of the highest resource category.
Data Source
AI summary
Techniques for optimizing the sampling of both the highest resource source category and lower resource source categories of a given categorical source data set with an initial distribution such that a target distribution of the given categorical source data set may be reached are described. Data items of the respective categories may be indexed and sampled such that the number of data items that are sampled more than once are tracked. Sampling, according to the target distribution, may continue until stop criteria are satisfied. In some cases, the respective indexes are used to determine the moment at which the highest resource source category is fully sampled, therefore minimizing the number of duplicate data items and optimizing the use of the respective categories of data items. The target distribution may then be used to train a machine learning model.