概率统计讨论班——金融机器学习模型的风险管理 Risk management for machine learning models in finance

报告人：陈当行（助理教授，昆山杜克大学）

时间：6月11日下午2：00-3：00

地点：海纳苑2幢102

摘要：近年来，机器学习模型在许多领域取得了巨大成功。尽管与传统的数学统计模型相比，机器学习模型在精度上取得了极大提高，然而对于诸如金融等高风险领域，精度并不是唯一的考量标准。模型的透明性、可解释性、合理性以及公平性对于金融监管也至关重要。如果无法满足监管需求，即便是高精度的机器学习模型也难以被业界广泛使用。在这个工作中，在金融知识的背景下，我们考虑了三类不同的单调性。在单调性的基础上，我们提出了一类新的精确、透明、合理且公平的神经网络模型。

Abstract:Machine learning methods have been very successful in a variety of fields, including computer vision and natural language processing. Even though machine learning models significantly improve accuracy over traditional mathematical and statistical models, accuracy is not the only concern when it comes to highly-staked sectors, such as the financial industry. Transparency, explainability, conceptual soundness, and fairness are at least as important as accuracy in the highly regulated finance sector. Without meeting regulatory requirements, even highly accurate machine learning methods are unlikely to be accepted. In this work, we present three types of monotonicity and propose a new family of neural networks that are accurate, transparent, conceptually sound, and fair.

联系人：林智（linzhi80@zju.edu.cn）