Project Description
The objective is to develop interpretable machine-learning classification algorithms from black-box classification algorithms. Many commonly used machine-learning methods, like neural networks, are very good at the tasks they were designed to perform, like classification (a generic term that includes e.g. image recognition). However, they are essentially black boxes for the end user. This is a shortcoming in certain fields, like medicine, where both doctors and patients need to understand the factors behind the algorithm's decisions. In this project we will develop interpretable machine learning algorithms, specifically classification trees, from black-box neural networks, using pseudo-samples generated in the data space.
Tasks and Responsibilites
The student will implement the proposed methodology in the language programming R and will run a few real-data examples, using benchmark data sets from the UCI Machine Learning Repository.
Desired Qualifications
None Listed.