With the dissemination of real-life Artificial Intelligence applications, validation of applied models becomes a fresh area to be standardized. Model validation has been for many years the responsibility of credit risk modelers, more familiar with regression-based models developed on tabular data. Nowadays, we observe several AI models using different neural network architectures and divergent algorithms of machine learning on not only tabular data but also text and image data.
Within the AI4Risk Expert Circle project, in this 45-minute live session we will discuss Prometeia’s AI-based Model Validation Framework related to validating machine learning and even deep learning models on four distinctive pillars: Process, Data, Governance and Method. We will focus on how these corresponding pillars serve to two different spectrums of AI model validation: qualitative (discriminative power) and quantitative (calibration power).
One subject matter expert on credit risk model validation (Cem Arısoy, DenizBank) will ask about the difficulties of validating such an extended scope of algorithms. As Prometeia, we will deep dive into those pillars and demonstrate the importance of such an end-to-end validation framework. Our aim is not to bridge the gap for the validation of popular algorithms (like xgboost), in fact to establish an all-time applicable framework which also guide on the wide-range of possible AI applications.
Introduction to AI-based Projects and AI Model Validation: Prometeia’s main focus
Fireside Chat: A discussion on Prometeia’s AI Model Validation Framework
Cem Arısoy, CFA, FRM
SVP, Credit Risk Control and Risk Models Validation, DenizBank
Senior Manager, Prometeia