Machine Learning Model Training Data Filtering via Faulty Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence systems face challenges in distinguishing between high-quality and low-quality data, leading to inaccurate model training and potentially harmful decisions, such as increased cybersecurity incidents due to non-representative training data.
Innovation Solution
A computing system identifies a 'faulty feature' in a dataset to differentiate between data representative and non-representative of a target population, generating a training dataset with samples that are more likely to be members of the target population, thereby improving data quality and model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If all available data is used for training, then the quantity of training data is maximized, but the quality and representativeness of the training data deteriorates due to inclusion of low-quality non-representative samples
Solution Approach 1:
The patent segments the training dataset by identifying and separating samples based on the faulty feature value. Samples are divided into representative and non-representative groups, allowing the system to selectively use only the representative portion for training, thus resolving the contradiction between quantity and quality of training data.
Solution Approach 2:
The patent applies local quality by making the training dataset heterogeneous in terms of selection criteria. Instead of uniformly including all data, the system selectively includes samples with specific faulty feature values that are representative of the target population, thereby improving local quality while maintaining adequate quantity.
2Quantity of substance
If low-quality non-representative data is used for training, then the quantity of training data is sufficient, but the model accuracy deteriorates due to misleading patterns learned from poor quality data
Solution Approach 1:
The patent segments the training data to isolate and exclude non-representative samples that would teach misleading patterns. By separating samples based on the faulty feature, the system ensures that only high-quality representative data is used, preventing the model from learning incorrect patterns while maintaining sufficient training quantity.
Solution Approach 2:
The patent extracts and removes the harmful element (non-representative samples with problematic faulty feature values) from the training dataset. This extraction process eliminates the source of misleading patterns, allowing the model to achieve higher accuracy on representative data alone.
3Reliability
If a faulty feature is identified and used to filter data, then the representativeness of training data is improved, but the device complexity increases due to additional data processing steps
Solution Approach 1:
The patent extracts the faulty feature from the dataset and uses it as a simple filtering criterion. This approach improves representativeness by removing non-representative samples while avoiding complex processing, as the solution relies on identifying and removing a specific feature rather than implementing complex data processing pipelines.
Solution Approach 2:
The patent changes the parameter of data selection by introducing a filter based on the faulty feature value. This simple parameter-based filtering approach improves data representativeness without significantly increasing system complexity, as it involves straightforward conditional logic rather than complex processing mechanisms.
4Measurement precision
If comprehensive data analysis is performed to identify representative samples, then the measurement precision of data quality assessment is improved, but the time required for data processing increases
Solution Approach 1:
The patent extracts the faulty feature as a key indicator of data representativeness and uses it as the primary criterion for sample selection. This extraction approach enables precise data quality assessment by focusing on a specific distinguishing feature, thereby achieving high measurement precision without requiring time-consuming comprehensive analysis of all data attributes.
Data Source
AI summary
In some aspects, a computing system may identify a feature that can be used to distinguish between data that is more likely to be representative of the target population. A computing system may identify a feature in a dataset where a first value of the feature is associated with a higher likelihood that a corresponding sample is not a member of the target population. Due to the differences between samples that have the first value and samples that have the second value, the computing system may determine that samples with the first value are less likely to be members of the target population or samples with the second value are more likely to be members of the target population. The computing system may determine that a training dataset should be generated using samples that have the second value.