Fellowship Final Report: Quantifying Machine Learning Model Trustworthiness

The title of the proposed project was “Quantifying Machine Learning Model Trustworthiness.” Machine learning (ML) algorithms are at the core of modern AI. The focus of ML researchers in the past decade has been mainly on improving the performance of ML by developing novel algorithms to outperform the prior algorithms and complete even more complex tasks. The enthusiasm in the ML research community combined with the thirst of the ML consumers (e.g., self-driving car manufacturers) pushed the ML community to speed up the deployment phase of the developed algorithms. This fast-paced process of development and deployment resulted in a phenomenon called by Google researchers as technical debt for ML systems.

More from the Catalyst