Making ML Model Error an Information Source

Explore a novel approach to machine learning model error analysis in this 34-minute conference talk from code::dive 2023. Discover how to transform model errors into valuable information sources, particularly in the context of mobile network industry applications. Learn techniques for assessing configuration changes and their impacts on performance indicators by analyzing model residuals. Gain insights into defining measures that reveal network performance gains or losses based on error analysis. Presented by Marta Hendler, a Data Scientist at Nokia and PhD student in biomedical engineering, this talk challenges conventional thinking about ML model errors and demonstrates their potential as tools for deriving useful system state information.