題目:Machine Learning for Disease Risk Stratification, Online Adverse Events Extraction and Financial Credit Scoring

主講人：周建棟

時間：2021年6月15日15：00-18:00

地點：bwin必贏唯一官網(wǎng)315會議室

騰訊會議號：400 583 858

講座簡介：

Recent investigations of novel machine learning techniques were developed for individualized disease risk stratification, adverse events extraction and financial credit scoring will be presented. Firstly, we incorporated latent patterns between clinical characteristics of patients with Brugada Syndrome using nonnegative matrix factorization to improve risk stratification of severe arrhythmic outcomes. Secondly, we developed a novel semantic and sentiment enriched multi-task bidirectional LSTM deep learning model (S2-Multitask-BiLSTM) to extract marijuana-related adverse events from massive social media texts for public health safety surveillance. Thirdly, we proposed an intrinsically interpretable factorization machine model for credit scoring. The model is able to capture the nonlinear interaction patterns among features, handle data sparsity, and efficiently deal with data imbalance issues with an asymmetric non-convex                                                -loss function. Interpretation with visualizations provides risk management agents with useful tools to observe the prediction strengths of both individual features and pair-wise feature interactions.

主講人簡介:

    周建棟，香港城市大學數(shù)據(jù)科學學院數(shù)據(jù)科學專業(yè)在讀博士，現(xiàn)階段主要研究方向包括可解釋人工智能，計算醫(yī)療和金融科技等。以第一作者(或共同第一作者)身份在Gut, IEEE Transactions on Fuzzy Systems, Transportation Research Part B: Methodological, Information Sciences, Journal of the American Heart Association, Journal of Hypertension, Pharmacological Research, European Journal of Clinical Investigation, Heart Rhythm, Frontiers in Physiology等SCI/SSCI JCR Q1國際主流期刊上發(fā)表論文15余篇，同時有多篇論文在INFORMS Journal on Computing，European Journal of Operational Research, Industrial Management & Data Systems, Information Sciences, Cardiovascular Drugs and Therapy等期刊審稿及獲得返修意見，曾多次在國際高水平會議做論文報告。