Fazl Barez

Research project title: 
Constrained Machine Learning for Interpretability and Robustness
About me: 

As the use of autonomous systems and "black-box" algorithms increases, the ability for these mechanisms to explain their output becomes increasingly important. Despite the unprecedented success of Machine Learning algorithms in tasks ranging from self-driving cars, surgical robots, film and book recommendations, to mortgage qualification, there is still no consensus on how their outputs are determined. However, in life-changing scenarios such as self-driving cars and disease diagnoses it is important to know the reasons behind such critical decisions. This project will develop a principle model for interpretability and robustness via the use of constraints.