Type: Bachelor Master UROP
Status: Available
The recent changes related to the EU’s General Data Protection Regulation (“GDPR”)[1] tasked the machine-learning (ML) community to enable explainability of the models and their output. More specifically, according to the GDPR, the ML models should offer the possibility to answer/provide explanation such as: “You were denied a loan because your annual income was £30,000. If your income had been £45,000, you would have been offered a loan.” [1]. This new regulation probed the ML community to explore the explainability of ML models to the extent that some researchers argue that model’s accuracy should be sacrificed and interpretable models should be preferred over black-box ML models for high-stake decisions [2]. In the ML domain, this task is referred to as search for counterfactual explanations. The idea is that, besides the model’s output, additional counterfactual information should be provided of how the world would have to be different for a desirable outcome to occur.
In the context of a BSc project, you will: (i) overview the related work on Explainable AI (XAI) and counterfactual explanations (starting from [3-9]); (ii) overview benchmark datasets for XAI; (iii) implement at least one existing XAI method and analyze its pros and cons on several datasets; (iv) summarize the results and propose future work.
In the context of a MSc project, you will: (i) overview the related work on Explainable AI (XAI) and counterfactual explanations (starting from [3-9]); (ii) overview benchmark datasets for XAI; (iii) implement at least one existing XAI method and analyze its pros and cons on several datasets; (iv) develop an improved XAI method and analyze its pros and cons WRT to the implemented method in (iii) on several datasets; (v) summarize the results and propose future work.
References
[1] Wachter, S., Brent M., and Chris R. Counterfactual explanations without opening the black box: Automated decisions and the GDPR. Harv. JL & Tech. 31 : 841, 2017.
[2] Rudin, C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence 1, no. 5:206-215, 2019.
[3] Rudin, C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence 1, no. 5:206-215, 2019.
[4] Joshi, S., Oluwasanmi K., Warut V., Been K., and Joydeep G. Towards realistic individual recourse and actionable explanations in black-box decision making systems. arXiv preprint arXiv:1907.09615, 2019.
[5] Karimi, A.-H., Gilles B., Borja B., and Isabel V. Model-agnostic counterfactual explanations for consequential decisions. arXiv preprint arXiv:1905.11190, 2019.
[6] Wexler, J., Pushkarna, M., Bolukbasi, T., Wattenberg, M., Viégas, F., and Wilson, J. The what-if tool: Interactive probing of machine learning models. IEEE transactions on visualization and computer graphics, 26(1):56–65, 2019.
[7] Tolomei, G., Silvestri, F., Haines, A., and Lalmas, M. Interpretable predictions of tree-based ensembles via actionable feature tweaking. In Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 465–474. ACM, 2017.
[8] Ustun, B., Spangher, A., and Liu, Y. Actionable recourse in linear classification. In Proceedings of the Conference on Fairness, Accountability, and Transparency, pages 10–19. ACM, 2019.
[9] Dandl, Susanne, Christoph Molnar, Martin Binder, and Bernd Bischl. “Multi-objective counterfactual explanations.” In International Conference on Parallel Problem Solving from Nature, pp. 448-469. Springer, Cham, 2020.
For more information contact: