Example-based Explanation in Machine Learning (co-supervised with Umang Bhatt) – Part II
Interpretability and explanation of machine learning algorithms is getting increasing attention. Example-based explanation is a specific type of explanation that aims at answering which data point in the training dataset is most influential to the model’s output for a specific test data point. The aim of this project is to implement four approaches to this form of explanation: Influence functions , Sequential Bayesian Quadrature , RelatIF , and Representer Points . The approaches will be compared using datasets such as CIFAR-10, AwA, MNIST, Enrol Spam Email, COMPAS, and/or ChemReact.
The ideal candidate for this project will have a strong background in mathematics and great Python programming skills. Familiarity with explanatory AI is a big plus.
 https://arxiv.org/abs/1703.04730 (ICML 2017)
 https://arxiv.org/abs/1810.10118v1 (AISTATS 2019)
 https://arxiv.org/abs/2003.11630 (AISTATS 2020)
 https://arxiv.org/abs/1811.09720 (NeurIPS 2018)
A Tool for Rule Extraction Methods (REM) – PartIII/MPhil
Ensemble and deep learning models are increasingly used in clinical decision-making, but the reasons for their decisions are often obscure. Rule-based models, common in classic expert systems, on the other hand, are transparent and can be used for simulation, thus facilitating easy inspection for clinicians. The aim of this project is to develop a clinical decision support system (CDSS) tool for cancer. At the backend the CDSS operates based on a ruleset extracted from neural network or tree-based approaches (random forest and decision tree). The frontend will accommodate functionalities such as prediction and explanation by the ruleset, filtering of the rules based on clinicians domain knowledge and visulaisation of the rules. In order to ensure that the tool is usable by clinicians potential user studies may be conducted.
The idea candidate for this project will have a strong background in programming (Python and Java Scripts) and logic. Machine learning experience and experience with health data is a big plus.