Data Science London MeetUp

Machine learning is increasingly being used in high-risk settings such as healthcare, finance and autonomous systems. However, there are often no assurances that these models are trustworthy: they may be overly confident in their predictions and may lack interpretability.

Conformal prediction offers a powerful framework for quantifying the uncertainty of model predictions in a statistically rigorous way. In this talk, we’ll explore the theory and demonstrate range of practical applications of conformal prediction, showing how it can help build end users' confidence in ML systems. In addition, Aman will present his current research on leveraging conformal prediction to aid with algorithmic recourse and explainability through uncertainty-aware counterfactual explanations.

Speaker:
Aman Bilkhoo is a researcher at the Data-Driven Verification lab at the Department of Informatics, King's College London. His research interests include explainability, privacy and uncertainty quantification in machine learning. His current work is on the application of conformal prediction to generating uncertainty-aware counterfactual explanations. Aman has also engineered AI systems for public sector clients and has previously worked on time-series conformal prediction with financial applications as part of his MSci, also obtained at King's College London.

Please register here - https://forms.office.com/Pages/ResponsePage.aspx?id=DbZwxsfE7E2MH6wNo-cr7ocfDRIwSD1IkQJYjHSNY3VUMFM4WENJODdWR1hHOFFESkFKNENXUzJPMS4u

By registering for this in person event you are agreeing to have your e-mail shared with building security for building access. Your privacy is important to us. This privacy statement explains the personal data Microsoft processes, how Microsoft processes it, and for what purposes. You can view our full privacy statement here:
https://www.microsoft.com/en-gb/privacy/privacystatement