BEGIN:VCALENDAR VERSION:2.0 X-WR-CALNAME:EventsCalendar PRODID:-//hacksw/handcal//NONSGML v1.0//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York LAST-MODIFIED:20240422T053451Z TZURL:https://www.tzurl.org/zoneinfo-outlook/America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZNAME:EDT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZNAME:EST TZOFFSETFROM:-0400 TZOFFSETTO:-0500 DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT CATEGORIES:College of Engineering,Thesis/Dissertations DESCRIPTION:EAS Doctoral Dissertation Defense by by Guancheng Zhou Date: We dnesday May 6, 2026 Time: 10:30am Topic: Towards Diagnosis, Fairness, and interpretation of Machine Learning Algorithms Location: Library 314 Abstra ct: A number of machine learning algorithms have delivered superior empiri cal performance. However, the understanding of their mechanisms has been h ampered by the black-box nature of the algorithms. In this proposal, we ap proach the problem from two diﬀerent lens. One is visualization, with a data-driven geometry following kernel—the rpf-kernel, which can extract complex and highly nonlinear patterns beyond the usual principal component analysis. The other is the diagnosis perspective. Specifically, we perfor m a diagnostic analysis to data points under a given algorithm and hope to use this as a proxy to understand the algorithm. Random Forests classific ation is used as an example algorithm for our study. We borrow two metrics , leverage and influence, from statistics regression to measure the import ance of data points, while extending their definition to a small neighborh ood of data points. Also studied is a related issue of fairness—whether the algorithm delivers a response that is fair in terms of some given metr ic, for example the gender of the associated subjects. K-means clustering is studied, and a computational eﬃcient post-algorithm adjustment method is proposed. Experiments show that the proposed method is eﬀective in i mproving the fairness while maintaining the clustering performance. Variab le importance is of major significance in the practice of statistical anal ysis and model interpretation. However, current methods do not consider th e correlation between variables, we proposed a method to solve this proble m and obtained a more reasonable variable importance. ADVISOR(S): Dr. Do nghui Yan, Department of Mathematics (dyan@umassd.edu) COMMITTEE MEMBERS: Dr. Haiping Xu, Department of Computer& Information Science Dr. Hongkang Xu, Department of Accounting & Finance Dr. Long Jiao, Department of Comput er & Information Science NOTE: All EAS Students are ENCOURAGED to atten d.\nEvent page: https://www.umassd.edu/events/cms/eas-doctoral-dissertatio n-defense-by-guancheng-zhou.php X-ALT-DESC;FMTTYPE=text/html:

EAS Doctoral Dissertation Defen se by

by Guancheng Zhou

Date: Wednesday May 6\, 2026

Time: 10:30am

Topic: Towards Diagnosis\, Fairness\, and interpr etation of Machine Learning Algorithms

Location: Library 314

Abstract:

A number of machine learning algorithms have delivere d superior empirical performance. However\, the understanding of their mec hanisms has been hampered by the black-box nature of the algorithms. In th is proposal\, we approach the problem from two diﬀerent lens. One is vis ualization\, with a data-driven geometry following kernel—the rpf-kernel \, which can extract complex and highly nonlinear patterns beyond the usua l principal component analysis. The other is the diagnosis perspective. Sp ecifically\, we perform a diagnostic analysis to data points under a given algorithm and hope to use this as a proxy to understand the algorithm. Ra ndom Forests classification is used as an example algorithm for our study. We borrow two metrics\, leverage and influence\, from statistics regressi on to measure the importance of data points\, while extending their defini tion to a small neighborhood of data points. Also studied is a related iss ue of fairness—whether the algorithm delivers a response that is fair in terms of some given metric\, for example the gender of the associated sub jects. K-means clustering is studied\, and a computational eﬃcient post- algorithm adjustment method is proposed. Experiments show that the propose d method is eﬀective in improving the fairness while maintaining the clu stering performance. Variable importance is of major significance in the p ractice of statistical analysis and model interpretation. However\, curren t methods do not consider the correlation between variables\, we proposed a method to solve this problem and obtained a more reasonable variable imp ortance.

ADVISOR(S): Dr. Donghui Yan\, Department of Mathematics

(dyan@umassd.edu)

COMMITTEE MEMBERS:

Dr. H aiping Xu\, Department of Computer& Information Science
Dr. Hong kang Xu\, Department of Accounting & Finance
Dr. Long Jiao\, Dep artment of Computer & Information Science

NOTE: All EAS Students are ENCOURAGED to attend.

Event page: https://www.umassd.edu/events/cms/eas-doctoral-dissertation-defense-b y-guancheng-zhou.php

DTSTAMP:20260410T163923 DTSTART;TZID=America/New_York:20260506T103000 DTEND;TZID=America/New_York:20260506T123000 LOCATION:LIB 314 SUMMARY;LANGUAGE=en-us:EAS Doctoral Dissertation Defense by Guancheng Zhou UID:3cbbbf08a6904576a51d69b4a0eb1a55@www.umassd.edu END:VEVENT END:VCALENDAR