Automated Dataset Reduction via Explainability Pathways
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The process of training machine learning models is time-consuming and resource-intensive, requiring large datasets where some data may negatively impact model accuracy, making it challenging to identify and exclude irrelevant data without reducing model performance.
Innovation Solution
An automated method using explainability techniques to identify pathways in machine learning models that decrease accuracy, correlating these pathways with specific datapoints, and removing them to generate a reduced training dataset that maintains model accuracy, thereby reducing dataset size and resource requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a large training dataset is used to train the machine learning model, then the model accuracy is improved, but the training time and computational resources increase significantly
Solution Approach 1:
The patent extracts and removes harmful datapoints from the training dataset that negatively impact model accuracy. By using explainability techniques to identify and extract only the beneficial datapoints, the system creates a reduced training dataset that maintains model accuracy while reducing training time and computational resource requirements
Solution Approach 2:
The patent changes the composition parameter of the training dataset by removing specific harmful datapoints identified through explainability analysis. This parameter change in dataset composition allows the model to achieve the same accuracy with fewer data points, thereby reducing training time and resource consumption
2Reliability
If a large training dataset is used to train the machine learning model, then the model accuracy is improved, but the storage space requirements increase
Solution Approach 1:
The patent extracts and removes harmful datapoints from the training dataset that negatively impact model accuracy. By using explainability techniques to identify and extract only the beneficial datapoints, the system creates a reduced training dataset that maintains model accuracy while reducing storage space requirements
3Reliability
If manual review processes are used to identify relevant training data, then the quality of training data is improved, but the process becomes time-consuming and resource-intensive
Solution Approach 1:
The patent implements an automated system that uses explainability techniques to self-identify harmful datapoints and their correlation with model pathways. This automated self-service approach eliminates the need for manual review processes, significantly improving data processing efficiency while maintaining training data quality through systematic identification of harmful patterns
Solution Approach 2:
The patent uses explainability techniques to provide feedback about how specific datapoints affect model pathways and accuracy. This automated feedback mechanism identifies harmful datapoints and their impact on model performance, enabling efficient automated dataset reduction without manual intervention
Data Source
AI summary
Systems, methods, and apparatuses for automatically generating reduced training datasets are described. A training dataset may be inputted into a machine learning model to train the machine learning model to output a label. The machine learning model may comprise nodes, and each of the nodes may be associated with a weight. Based on datapoints, changes to the weight associated with each node of the plurality of nodes may be determined. Using model explainability techniques and based on the changes to the weight associated with each node of the plurality of nodes, pathways that decrease an accuracy of the machine learning model are identified. A first set of the datapoints that correlate with pathways that decrease the accuracy of the machine learning model outputting the label may be determined. Furthermore, the first set of the datapoints may be removed from the training dataset to generate a reduced training dataset.