Insights into Fairness through Trust: Multi-scale Trust Quantification for Financial Deep Learning

The success of deep learning in recent years have led to a significant\nincrease in interest and prevalence for its adoption to tackle financial\nservices tasks. One particular question that often arises as a barrier to\nadopting deep learning for financial services is whether the developed\nfinancial deep learning models are fair in their predictions, particularly in\nlight of strong governance and regulatory compliance requirements in the\nfinancial services industry. A fundamental aspect of fairness that has not been\nexplored in financial deep learning is the concept of trust, whose variations\nmay point to an egocentric view of fairness and thus provide insights into the\nfairness of models. In this study we explore the feasibility and utility of a\nmulti-scale trust quantification strategy to gain insights into the fairness of\na financial deep learning model, particularly under different scenarios at\ndifferent scales. More specifically, we conduct multi-scale trust\nquantification on a deep neural network for the purpose of credit card default\nprediction to study: 1) the overall trustworthiness of the model 2) the trust\nlevel under all possible prediction-truth relationships, 3) the trust level\nacross the spectrum of possible predictions, 4) the trust level across\ndifferent demographic groups (e.g., age, gender, and education), and 5)\ndistribution of overall trust for an individual prediction scenario. The\ninsights for this proof-of-concept study demonstrate that such a multi-scale\ntrust quantification strategy may be helpful for data scientists and regulators\nin financial services as part of the verification and certification of\nfinancial deep learning solutions to gain insights into fairness and trust of\nthese solutions.\n