Model validation and explainability

The event is postponed to March 2023, we will inform you about the new date soon.

Summary

Striking the right balance between performance and explainability can present a difficult dilemma for financial institutions. By making some well-targeted enhancements (i.e. focus on interpretability, bias, feature engineering, hyperparameter tuning, etc.), existing validation frameworks can address machine-learning-model risks and help companies gain the confidence to start harnessing the full power of machine learning.

Topic 1: De-risking machine learning in financial services

Overview: The US Fed SR 11-7 paper, published in 2011, and the TRIM Guide for Internal Models published by the European Central Bank in 2017, broadly outline the role of model validators in assessing models from four perspectives: conceptual soundness, process verification, ongoing monitoring and outcomes analysis. When used on their own, however, these frameworks are insufficient in assessing the full scope of risks ML models may pose. Some of the reasons for this include the use of alternative and vastly larger datasets, increased complexity of ML models compared to their statistical counterparts and the need for more specialized skills.

Topic 2: Towards explainable AI in the Financial sector

Striking the right balance between performance and explainability can present a difficult dilemma for financial institutions. This session will focus on the challenges associated with the former statement, the need for differentiation between technical explainability requirements and explainability requirements of the broader AI systems in relation to laws and regulations, and different techniques bank can use to activate explainability (e.g. simple local analysis, deep model analysis, surrogate model development, contrastive analysis).

Presenters

Snezhana Ilieva: Lead Data Engineer at EPAM
moderated by Anton Seidel.

Agenda

18:00 Opening (Anton Seidel)
18:10 De-risking machine learning in financial services (Snezhana Ilieva)
18:35 Towards explainable AI in the Financial sector (Dean Palejev)
19:00 Q&A
19:30 Networking drinks (all)
20:30 end

Chapter Event

In our chapter events – access for members only – we present one or more speakers to share knowledge, updates and best practices on a specific risk topic. In small groups of risk professionals you can exchange thoughts and test ideas. More on SRA chapters. This event is hosted by the chapter Risk Analytics and Models.