Publications by authors named "Ronald G Askin"
A Simulation-based Approach for Quantifying the Impact of Interactive Label Correction for Machine Learning.
IEEE Trans Vis Comput Graph
September 2024
Article Synopsis
- Recent studies have focused on how the characteristics of training data impact the performance of machine learning models, particularly the value of correcting labels through human interaction.
- Limited research has quantitatively assessed the cost-benefit relationship of label correction in terms of performance improvement across various conditions, like datasets and algorithms.
- Using simulations, the researchers found that while label correction can boost model performance, its effectiveness varies based on task conditions, leading to recommendations for practitioners on when it is most beneficial to use interactive label correction.