Facebook AI is seeking an alternative approach to ensure fairness in decisions made by machine learning systems. The tech giant explores a distinct method to apply algorithmic fairness techniques to complex and large-scale production systems.
In a research paper dubbed “Fairness on the Ground: Applying Algorithmic Fairness Approaches to Production Systems,” the Company reconnoiters the expertise of integrating fairness applications into complex and large-scale production systems meant to benefit other disciplines facing similar problems.
The research conducted by Facebook AI construes statistical measures of fairness such as calibration and equality of odds to reasonable practical approach. The research isn’t the first one to identify the challenges facing complex systems.
According to Facebook AI team, “the challenges of bias in supervised machine learning models are most likely to be studied problem in algorithmic fairness.”
In conclusion, the study presented a holistic approach to addressing problems developed within the context of large technology company. Fairness was considered at three levels, which include product, policy and implementation.
The study was able to distinguish between normative questions from statistical ones where applicable. It presented a high-level approach, at the implementation level, to train fairness questions based on costs and benefits produced by decisions.
The typical binary decision-making context was classified into algorithmic dcesion making and human labeling.
Comments
Post a Comment