16 Dec 2020
Managing Quality and Explainability in Machine Learning and AI
In an increasingly data-driven world both the data being available to actuaries in their daily work as well as the toolbox of algorithms grow significantly. This is because new sources of data are open for actuarial work, like geo-risk data on granular levels, data from sensors and all kinds of IoT devices, health trackers, telematics data etc. On the algorithmic side, methods from the area of analytics, predictive modelling and machine learning as well applications in the area of artificial intelligence (ML in the following). These methods also extend the field of activity of actuaries, e.g. into the field of customer behaviour analytics.
Within this brave new world actuaries must maintain the highest professional standards of data and algorithmic quality. This means that there has to be a systematic management of input data as well as of algorithmic quality throughout actuarial procedures.
We start the web session with a look on the legal and regulatory side. What are the most relevant dimensions of data and algorithmic quality and how does this refer to explainability and fairness of ML? What are relevant measures of quality in this respect?
We proceed by having a look at existing European regulation and legislation, including actuarial quality standards, which govern the use and the application of these new methodologies. We also include current discussions which are underway in insurance, actuarial, regulatory and ML communities.
In order to take a view on the practical side, we show and discuss various approaches of creating transparency and of managing risk around ML. This refers to both the most relevant selection of ML approaches, but also to additional measures to interpret ML results (like LIME, SHAP etc.). We also refers to most useful R/Python libraries.
Finally, we will have a look at a typical use case, and present how key quality challenges can be tackled in this specific case. This should enable the participant to transfer practical measures of data and algorithmic quality management to his/her own realm.
Organised by the EAA - European Actuarial Academy GmbH.
The web session is open to all interested persons, who ideally (but not necessarily) have a basic knowledge of the algorithmic universe of ML.
Technical requirementsPlease check with your IT department if your firewall and computer settings support web session participation (the programme GotoTraining/GotoWebinar is used for the web session). Please also make sure that you are joining the web session with a stable internet connection.
Purpose and Nature