BEGIN:VCALENDAR PRODID:-//Microsoft Corporation//Outlook 16.0 MIMEDIR//EN VERSION:2.0 METHOD:PUBLISH X-MS-OLK-FORCEINSPECTOROPEN:TRUE BEGIN:VTIMEZONE TZID:Central Standard Time BEGIN:STANDARD DTSTART:16011104T020000 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZOFFSETFROM:-0500 TZOFFSETTO:-0600 END:STANDARD BEGIN:DAYLIGHT DTSTART:16010311T020000 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZOFFSETFROM:-0600 TZOFFSETTO:-0500 END:DAYLIGHT END:VTIMEZONE BEGIN:VEVENT CLASS:PUBLIC CREATED:20220912T205237Z DESCRIPTION:Speaker:\nArlei Silva - Assistant Professor of Computer Science\, Rice University\n\nTit le:\nLink Prediction with Autocovariance\n\nAbstract:\nMachine learning on graphs supports various structured-data applications including social net work analysis\, recommender systems\, and natural language processing. One could argue that link prediction is the most fundamental among the graph- related tasks. This is because link prediction not only has many concrete applications (e.g. friendship and product recommendation\, uncovering prot ein-protein interactions) but can also be considered an (implicit or expli cit) step of most graph-based machine learning pipelines due to the fact t hat the observed graph is often incomplete. Earlier link prediction approa ches relied on expert-designed heuristics (e.g.\, Common Neighbors\, Adami c-Adar\, Preferential Attachment) to extract topological information from the network. More recently\, representation learning on graphs and Graph N eural Networks (GNNs) have emerged as the predominant solutions for link p rediction.\nIn this talk\, we will introduce link prediction methods based on autocovariance\, which is a multiscale random-walk-based node similari ty metric. We will show that the proposed approaches achieve state-of-the- art performance on simple\, signed\, and attributed graphs. As some of our key findings\, we show that representation learning results for node clas sification do not generalize to link prediction. Moreover\, autocovariance is especially accurate at predicting negative links in polarized signed g raphs. Finally\, our results illustrate how existing approaches for traini ng and evaluation of supervised link prediction\, including those based on GNNs\, picture an overly optimistic picture of their performance. We show that a simple approach combining autocovariance and attribute information outperforms several recent GNN-based link prediction methods.\nSpeaker Bi o:\nArlei Silva is an Assistant Professor of Computer Science at Rice Univ ersity. His research focuses on developing algorithms and models for minin g and learning from complex datasets\, broadly defined as data science\, e specially for data represented as graphs/networks. He is particularly inte rested in problems motivated by computational social science\, infrastruct ure\, and healthcare. The tools that he applies to address these problems include machine learning\, network science\, graph theory\, linear algebra \, optimization\, and statistics. Professor Silva received a Ph.D in Compu ter Science from the University of California\, Santa Barbara\, advised by Ambuj Singh\, where he was also a postdoctoral scholar.\nLocation:\nIn pe rson: Chambers Hall\, 600 Foster Street\, Lower Level\nRemote option: http s://northwestern.zoom.us/j/98973037019 \nPasscode: NICO22\nAbout t he Speaker Series:\nWednesdays@NICO is a vibrant weekly seminar series foc using broadly on the topics of complex systems and data science. It brings together attendees ranging from graduate students to senior faculty who s pan all of the schools across Northwestern\, from applied math to sociolog y to biology and every discipline in-between. Please visit: https://bit.ly /WedatNICO for information on future speakers.\n DTEND;TZID="Central Standard Time":20221005T130000 DTSTAMP:20220912T205237Z DTSTART;TZID="Central Standard Time":20221005T120000 LAST-MODIFIED:20220912T205237Z LOCATION:In person at Chambers Hall\, or remote via Zoom PRIORITY:5 SEQUENCE:0 SUMMARY;LANGUAGE=en-us:Wed@NICO\, 10/5\, Arlei Silva TRANSP:OPAQUE UID:040000008200E00074C5B7101A82E008000000007003837BBAC6D801000000000000000 010000000F89893804DCBFE41998CAD30C2201150 X-ALT-DESC;FMTTYPE=text/html:

Speaker:

Arlei Silva - Assistant Professor of Computer Science\, Rice University
< br>Title:

Link Prediction with Autocovariance
Abstract:

Machine learning on gr aphs supports various structured-data applications including social networ k analysis\, recommender systems\, and natural language processing. One co uld argue that link prediction is the most fundamental among the graph-rel ated tasks. This is because link prediction not only has many concrete app lications (e.g. friendship and product recommendation\, uncovering protein -protein interactions) but can also be considered an (implicit or explicit ) step of most graph-based machine learning pipelines due to the fact that the observed graph is often incomplete. Earlier link prediction approache s relied on expert-designed heuristics (e.g.\, Common Neighbors\, Adamic-A dar\, Preferential Attachment) to extract topological information from the network. More recently\, representation learning on graphs and Graph Neur al Networks (GNNs) have emerged as the predominant solutions for link pred iction.

In this talk\, we will intro duce link prediction methods based on autocovariance\, which is a multisca le random-walk-based node similarity metric. We will show that the propose d approaches achieve state-of-the-art performance on simple\, signed\, and attributed graphs. As some of our key findings\, we show that representat ion learning results for node classification do not generalize to link pre diction. Moreover\, autocovariance is especially accurate at predicting ne gative links in polarized signed graphs. Finally\, our results illustrate how existing approaches for training and evaluation of supervised link pre diction\, including those based on GNNs\, picture an overly optimistic pic ture of their performance. We show that a simple approach combining autoco variance and attribute information outperforms several recent GNN-based li nk prediction methods.

Speaker Bio:

Arlei Silva is an Assistant Professor of Computer Science at Rice University. His resear ch focuses on developing algorithms and models for mining and learning fro m complex datasets\, broadly defined as data science\, especially for data represented as graphs/networks. He is particularly interested in problems motivated by computational social science\, infrastructure\, and healthca re. The tools that he applies to address these problems include machine le arning\, network science\, graph theory\, linear algebra\, optimization\, and statistics. Professor Silva received a Ph.D in Computer Science from the University of California\, Santa Barbara\, ad vised by Ambuj Singh\, where he was also a postd octoral scholar.

Location:

In person: Cham bers Hall\, 600 Foster Street\, Lower Level
Remote option: https://northwestern.zoom.us/j/98973037019
Passcode: NIC O22

About the Speaker Series:

Wednesdays@NICO is a vibrant weekly seminar series focusing broadly on the topics of complex systems and data science. It brings together attendees ranging fro m graduate students to senior faculty who span all of the schools across N orthwestern\, from applied math to sociology to biology and every discipli ne in-between. Please visit: https://bi t.ly/WedatNICO for information on future speakers.

X-MICROSOFT-CDO-BUSYSTATUS:BUSY X-MICROSOFT-CDO-IMPORTANCE:1 X-MICROSOFT-DISALLOW-COUNTER:FALSE X-MS-OLK-AUTOFILLLOCATION:FALSE X-MS-OLK-CONFTYPE:0 BEGIN:VALARM TRIGGER:-PT15M ACTION:DISPLAY DESCRIPTION:Reminder END:VALARM END:VEVENT END:VCALENDAR