報告題目：Explainable Artificial Intelligence in Banking and Finance
報 告 人：Professor Aijun Zhang，Department of Statistics and Actuarial Science, The University of Hong Kong
報告時間：2019年12月26日，10:00—11:00
報告地點：數學樓一樓報告廳
報告摘要：
The recent developments of deep learning in artificial intelligence (AI) have brought great successes in computer vision and natural language processing. They are based on the deep neural networks that are too complex to interpret. Such black box AI models have limited applications in banking and finance. In this talk, a holistic view of interpretable machine learning will be presented, including post-hoc explainability and intrinsic modeling approaches from global and local interpretability perspectives. We suggest to enhance the intrinsic interpretability through a variety of model constraints. In particular, we propose a constructive approach to developing explainable neural networks through sparse, orthogonal and smooth constraints. We derive the necessary and sufficient identifiability conditions for the proposed model. The multiple parameters are simultaneously estimated by a modified mini-batch gradient descent method based on backpropagation algorithm and the Cayley transformation. By simulation and real case studies, the proposed explainable neural networks are shown to achieve the superior balance between prediction accuracy and model interpretability.
報告人簡介：
張愛軍博士現任香港大學統計及精算學系助理教授，主要從事大數據分析、大規模機器學習、可解釋人工智能等領域的基礎研究及其在銀行金融領域的實踐應用。他曾擔任香港大學“數據科學”碩士學位課程副主任和香港大學“應用人工智能”文理學士學位課程創始主任。加入香港大學前，張愛軍博士于2014至2016年在香港浸會大學深圳研究院擔任教育大數據中心主任，在任期間領導研究團隊成功開發新一代微慕課軟件和基于微課慕課的大規模教育數據挖掘平台；于2008至2013年任職美國銀行全球風險管理部，主要從事大規模信貸資産風險管理，參與發明美國專利“風險與獎勵評估機制”和多項量化風險模型研究。張愛軍博士于1998年考入清華大學，轉學香港浸會大學于2002年和2004年分别獲數學理學士學位和統計學哲學碩士學位，再赴美國密歇根大學于2009年獲統計學哲學博士學位。他目前擔任中國數學會均勻設計分會常務委員和廣東省高校統計學專業教學指導委員會委員。
