BEGIN:VCALENDAR PRODID:BedeWork V3.5 VERSION:2.0 METHOD:PUBLISH BEGIN:VTIMEZONE TZID:America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:20070311T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:20071104T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD BEGIN:STANDARD TZOFFSETFROM:-045602 TZOFFSETTO:-0500 TZNAME:EST DTSTART:18831118T120358 RDATE:18831118T120358 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:19180331T020000 RDATE:19180331T020000 RDATE:19190330T020000 RDATE:19200328T020000 RDATE:19210424T020000 RDATE:19220430T020000 RDATE:19230429T020000 RDATE:19240427T020000 RDATE:19250426T020000 RDATE:19260425T020000 RDATE:19270424T020000 RDATE:19280429T020000 RDATE:19290428T020000 RDATE:19300427T020000 RDATE:19310426T020000 RDATE:19320424T020000 RDATE:19330430T020000 RDATE:19340429T020000 RDATE:19350428T020000 RDATE:19360426T020000 RDATE:19370425T020000 RDATE:19380424T020000 RDATE:19390430T020000 RDATE:19400428T020000 RDATE:19410427T020000 RDATE:19460428T020000 RDATE:19470427T020000 RDATE:19480425T020000 RDATE:19490424T020000 RDATE:19500430T020000 RDATE:19510429T020000 RDATE:19520427T020000 RDATE:19530426T020000 RDATE:19540425T020000 RDATE:19550424T020000 RDATE:19560429T020000 RDATE:19570428T020000 RDATE:19580427T020000 RDATE:19590426T020000 RDATE:19600424T020000 RDATE:19610430T020000 RDATE:19620429T020000 RDATE:19630428T020000 RDATE:19640426T020000 RDATE:19650425T020000 RDATE:19660424T020000 RDATE:19670430T020000 RDATE:19680428T020000 RDATE:19690427T020000 RDATE:19700426T020000 RDATE:19710425T020000 RDATE:19720430T020000 RDATE:19730429T020000 RDATE:19740106T020000 RDATE:19750223T020000 RDATE:19760425T020000 RDATE:19770424T020000 RDATE:19780430T020000 RDATE:19790429T020000 RDATE:19800427T020000 RDATE:19810426T020000 RDATE:19820425T020000 RDATE:19830424T020000 RDATE:19840429T020000 RDATE:19850428T020000 RDATE:19860427T020000 RDATE:19870405T020000 RDATE:19880403T020000 RDATE:19890402T020000 RDATE:19900401T020000 RDATE:19910407T020000 RDATE:19920405T020000 RDATE:19930404T020000 RDATE:19940403T020000 RDATE:19950402T020000 RDATE:19960407T020000 RDATE:19970406T020000 RDATE:19980405T020000 RDATE:19990404T020000 RDATE:20000402T020000 RDATE:20010401T020000 RDATE:20020407T020000 RDATE:20030406T020000 RDATE:20040404T020000 RDATE:20050403T020000 RDATE:20060402T020000 END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:19181027T020000 RDATE:19181027T020000 RDATE:19191026T020000 RDATE:19201031T020000 RDATE:19210925T020000 RDATE:19220924T020000 RDATE:19230930T020000 RDATE:19240928T020000 RDATE:19250927T020000 RDATE:19260926T020000 RDATE:19270925T020000 RDATE:19280930T020000 RDATE:19290929T020000 RDATE:19300928T020000 RDATE:19310927T020000 RDATE:19320925T020000 RDATE:19330924T020000 RDATE:19340930T020000 RDATE:19350929T020000 RDATE:19360927T020000 RDATE:19370926T020000 RDATE:19380925T020000 RDATE:19390924T020000 RDATE:19400929T020000 RDATE:19410928T020000 RDATE:19450930T020000 RDATE:19460929T020000 RDATE:19470928T020000 RDATE:19480926T020000 RDATE:19490925T020000 RDATE:19500924T020000 RDATE:19510930T020000 RDATE:19520928T020000 RDATE:19530927T020000 RDATE:19540926T020000 RDATE:19551030T020000 RDATE:19561028T020000 RDATE:19571027T020000 RDATE:19581026T020000 RDATE:19591025T020000 RDATE:19601030T020000 RDATE:19611029T020000 RDATE:19621028T020000 RDATE:19631027T020000 RDATE:19641025T020000 RDATE:19651031T020000 RDATE:19661030T020000 RDATE:19671029T020000 RDATE:19681027T020000 RDATE:19691026T020000 RDATE:19701025T020000 RDATE:19711031T020000 RDATE:19721029T020000 RDATE:19731028T020000 RDATE:19741027T020000 RDATE:19751026T020000 RDATE:19761031T020000 RDATE:19771030T020000 RDATE:19781029T020000 RDATE:19791028T020000 RDATE:19801026T020000 RDATE:19811025T020000 RDATE:19821031T020000 RDATE:19831030T020000 RDATE:19841028T020000 RDATE:19851027T020000 RDATE:19861026T020000 RDATE:19871025T020000 RDATE:19881030T020000 RDATE:19891029T020000 RDATE:19901028T020000 RDATE:19911027T020000 RDATE:19921025T020000 RDATE:19931031T020000 RDATE:19941030T020000 RDATE:19951029T020000 RDATE:19961027T020000 RDATE:19971026T020000 RDATE:19981025T020000 RDATE:19991031T020000 RDATE:20001029T020000 RDATE:20011028T020000 RDATE:20021027T020000 RDATE:20031026T020000 RDATE:20041031T020000 RDATE:20051030T020000 RDATE:20061029T020000 END:STANDARD BEGIN:STANDARD TZOFFSETFROM:-0500 TZOFFSETTO:-0500 TZNAME:EST DTSTART:19200101T000000 RDATE:19200101T000000 RDATE:19420101T000000 RDATE:19460101T000000 RDATE:19670101T000000 END:STANDARD BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EWT DTSTART:19420209T020000 RDATE:19420209T020000 END:DAYLIGHT BEGIN:DAYLIGHT TZOFFSETFROM:-0400 TZOFFSETTO:-0400 TZNAME:EPT DTSTART:19450814T190000 RDATE:19450814T190000 END:DAYLIGHT END:VTIMEZONE BEGIN:VEVENT CATEGORIES:Engineering CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240129T183017Z DESCRIPTION:LUNCH: \nLunch will be served at 11:45 AM. \n\nABSTRACT: \nThe ever-increasing scale of foundation models\, such as ChatGPT and AlphaFo ld\, has revolutionized AI and science more generally. However\, increasi ng scale also steadily raises computational barriers\, blocking almost ev eryone from studying\, adapting\, or otherwise using these models for any thing beyond static API queries. In this talk\, I will present research t hat significantly lowers these barriers for a wide range of use cases\, i ncluding inference algorithms that are used to make predictions after tra ining\, fine-tuning approaches that adapt a trained model to new data\, a nd finally\, full training of foundation models from scratch. \n\nSPEAKER BIO: \nTim Dettmers' research focuses on making foundation models\, such as ChatGPT\, accessible to researchers and practitioners by reducing the ir resource requirements. A PhD candidate at the University of Washington who has won oral\, spotlight\, and best paper awards at conferences such as ICLR and NeurIPS\, Tim Dettmers created the bits-and-bytes library fo r efficient deep learning\, which is growing at 1.4 million installations per month and received Google Open Source and PyTorch Foundation awards. DURATION:PT1H DTSTAMP:20240129T183017Z DTSTART;TZID=America/New_York:20240212T120000 LAST-MODIFIED:20240129T183017Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Accessible Foundation Models: Systems\, Algorithms\, and Science UID:CAL-8a0292fd-8d13410f-018d-567c85b0-00004e26demobedework@mysite.edu URL:https://cs.duke.edu/events/accessible-foundation-models-systems-algori thms-and-science X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Engineering:/user/public-use r/Topics/Engineering X-BEDEWORK-SPEAKER:Tim Dettmers\, University of Washington X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=cynthia@cs.duke.edu:Cynt hia Rudin X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Tim Dettmers from University of Washington X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/timdettmersdukecalendar_20240129063017PM.j pg X-BEDEWORK-THUMB-IMAGE:/public/Images/timdettmersdukecalendar_202401290630 17PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Engineering CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240131T193228Z DESCRIPTION:LUNCH: \nLunch will be served at 11:45 AM. \n\nABSTRACT: \nEff iciency is increasingly tied to quality to machine learning\, with more e fficient training algorithms leading to more powerful models trained on m ore data. However\, today's most popular machine learning models are buil t on asymptotically inefficient primitives. For example\, attention in Tr ansformers scales quadratically in the input size\, which makes it challe nging for Transformers to use long context. In this talk\, I discuss my w ork on improving the efficiency of the core primitives in machine learnin g\, with an emphasis on hardware-aware algorithms and long-context applic ations. In the first half\, I focus on replacing attention with gated sta te space models (SSMs) and convolutions\, which scale sub-quadratically i n context length. I describe the H3 (Hungry Hungry Hippos) architecture\, a gated SSM architecture that matches Transformers in quality up to 3B p arameters and achieves 2.4x faster inference. In the second half\, I focu s on developing hardware-aware algorithms for SSMs and convolutions. I de scribe FlashFFTConv\, a fast algorithm for computing SSMs and convolution s on GPU by optimizing the Fast Fourier Transform (FFT). FlashFFTConv yie lds up to 7x speedup and 5x memory savings\, even over vendor solutions f rom Nvidia. FlashFFTConv is now widely used in many gated SSM models\, in cluding language models\, image generation models\, and long-context DNA foundation models. \n\nSPEAKER BIO: \nDan Fu is a PhD student in the Comp uter Science Department at Stanford University\, where he is co-advised b y Christopher Ré and Kayvon Fatahalian. His research interests are at the intersection of systems and machine learning. Recently\, he has focused on developing algorithms and architectures to make machine learning more efficient\, especially for enabling longer-context applications. His rese arch has appeared as oral and spotlight presentations at NeurIPS\, ICML\, and ICLR\, and he has received the best student paper runner up at UAI. Dan has also been supported by an NDSEG fellowship. DURATION:PT1H DTSTAMP:20240131T193228Z DTSTART;TZID=America/New_York:20240213T120000 LAST-MODIFIED:20240131T193228Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Hardware-Aware Efficient Primitives for Machine Learning UID:CAL-8a0292fd-8d13410f-018d-61022d84-000044c8demobedework@mysite.edu URL:https://cs.duke.edu/events/hardware-aware-efficient-primitives-machine -learning X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Engineering:/user/public-use r/Topics/Engineering X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-SPEAKER:Dan Fu\, Stanford University X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=cynthia@cs.duke.edu:Cynt hia Rudin X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Dan Fu of Stanford University X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/danfudukecalendar_20240131073228PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/danfudukecalendar_20240131073228PM-t humb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240131T171203Z DESCRIPTION:LUNCH: \nLunch will be served at 11:45 AM. \n\nABSTRACT: \nRec ent advancements in large language models (LLMs) have marked a significan t milestone in the field of natural language processing. Yet\, as we vent ure into the diverse and intricate terrain of the open world-encompassing varied topics\, domains\, and modalities-these models encounter formidab le challenges. Key among these are issues like hallucination\, grounded r easoning across multimodality\, and the burden of high computational dema nds. My research aims to tackle these challenges\, focusing on three broa d themes: (1) knowledge acquisition and understanding in the dynamic open world\; (2) generalizable and efficient intelligence\; (3) enhancing hum an experience when interacting with these technologies. In this talk\, I will mainly delve into the pursuit of generalizable and efficient intelli gence. I'll first introduce our latest advancements in endowing models wi th a broader cognitive scope\, evolving from answering simple questions t o complex questions\, and from understanding text to multimodality. Addit ionally\, I will also highlight our recent research on addressing task in terference\, a crucial but usually overlooked issue\, in the context of p arameter-efficient tuning. The talk will conclude with discussions into o ur future directions under these three pivotal themes. \n\nSPEAKER BIO: \ nLifu Huang is an Assistant Professor in the Computer Science department at Virginia Tech. He obtained a PhD in Computer Science from University o f Illinois at Urbana-Champaign in 2020. He has a wide range of research i nterests in natural language processing\, multimodal learning and machine learning. His research has been recognized with an Outstanding Paper Awa rd at ACL 2023 and Best Paper Award Honorable Mention at SIGIR 2023. He i s a recipient of the NSF CAREER Award in 2023 and Amazon Research Award i n 2021. DURATION:PT1H DTSTAMP:20240131T171203Z DTSTART;TZID=America/New_York:20240214T120000 LAST-MODIFIED:20240131T171203Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Bridging the Modern Language Modeling with the Complex Open World UID:CAL-8a0292fd-8d13410f-018d-60819e8e-00002ceedemobedework@mysite.edu URL:https://cs.duke.edu/events/bridging-modern-language-modeling-complex-o pen-world X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-SPEAKER:Lifu Huang X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=bdhingra@cs.duke.edu:Bhu wan Dhingra X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Lifu Huang X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/lifu-huangdukecalendar_20240131051203PM.jp g X-BEDEWORK-THUMB-IMAGE:/public/Images/lifu-huangdukecalendar_2024013105120 3PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240205T132521Z DESCRIPTION:LUNCH: \nLunch will be served at 11:45 AM. \n\nABSTRACT: \nAI has emerged as the most transformative and revolutionary technique\, resh aping many aspects of our lives. Its intersection with science\, particul arly physics\, has opened new avenues for understanding our world and uni verse. This understanding is grounded in centuries of exploration by bril liant minds. Physics studies today predominantly rely on rigorous methods founded on universal physical laws. I will discuss integrating advanced learning techniques\, notably Bayesian machine learning\, into computatio nal physics in this presentation. This integration is crucial in an inter disciplinary field that combines mathematics\, physics\, and computer sci ence to address meaningful\, real-world problems. As the first principle\ , physics offers novel techniques and insights for tackling complex tasks in complex\, structured data analysis. I envision synergizing physics an d probabilistic learning to create a formidable tool for exploring new fr ontiers. \n\nSPEAKER BIO: \nShibo Li\, a fifth-year Ph.D. candidate at th e University of Utah\, is affiliated with the Kahlert School of Computing (SoC) and the Scientific Computing and Imaging Institute (SCI). He earne d his master's degree from the University of Pittsburgh and his bachelor' s degree from the South China University of Technology. Shibo's research spans a range of topics\, including Bayesian machine learning\, approxima te inference\, interactive learning (encompassing Active Learning\, Bandi ts\, and Reinforcement Learning)\, and high-dimensional spatial-temporal modeling. His Ph.D. thesis focuses on multi-fidelity modeling and optimiz ation for physical simulations. His works have been published in top-tier machine learning and data mining avenues\, such as ICML\, NeurIPS\, ICLR \, AISTATS\, IJCAI\, and CIKM. DURATION:PT1H DTSTAMP:20240205T132521Z DTSTART;TZID=America/New_York:20240219T120000 LAST-MODIFIED:20240205T132521Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Physics-Motivated and Inspired Probabilistic Learning UID:CAL-8a0292fd-8d13410f-018d-7971dd18-00006640demobedework@mysite.edu URL:https://cs.duke.edu/events/physics-motivated-and-inspired-probabilisti c-learning X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-SPEAKER:Shibo Li\, University of Utah X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=tomasi@duke.edu:Carlo To masi X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Shibo Li\, University of Utah X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/shibolidukecalendar_20240205012521PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/shibolidukecalendar_20240205012521PM -thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240206T182839Z DESCRIPTION:LUNCH: \nLunch will be served at 11:45 AM. \n\nABSTRACT: \nFor mal logic programs are useful tools in AI. However\, they require users t o first express the problem in a formal logic language\, which is difficu lt to do for many real-world problems. In this talk\, I will discuss an a lternative paradigm\, using large language models (LLMs) as informal logi c programs. In this paradigm\, the propositions are expressed in natural language and the reasoning steps are carried out by a prompted LLM.\n\nTh is talk will present 3 problems effectively addressed by this paradigm: 1 .) Event sequence modeling and prediction\, the task of reasoning about f uture events given the past. 2.) Natural language entailment. 3.) Embodie d reasoning\, in which a robot needs to plan multiple steps to complete a task. For all these problems\, our paradigm achieves stronger results th an classical methods using formal logic programs and/or using LLMs as sta ndalone solvers. \n\nSPEAKER BIO: \nDr. Hongyuan Mei is currently a Resea rch Assistant Professor at Toyota Technological Institute at Chicago (TTI C). He obtained his PhD from Johns Hopkins University (JHU) Computer Scie nce. Hongyuan's research spans ML and natural language processing. Curren tly\, he is most interested in harnessing and improving the reasoning cap abilities of large language models to solve challenging problems such as event prediction. His research has been supported by a Bloomberg Data Sci ence PhD Fellowship\, the 2020 JHU Jelinek Memorial Award\, and research gifts from Adobe and Ant Group. His technical innovations have been integ rated into real-world products such as Alipay\, the world's largest mobil e digital payment platform\, which serves more than one billion users. Hi s research has been covered by Fortune Magazine and Tech At Bloomberg. DURATION:PT1H DTSTAMP:20240206T182839Z DTSTART;TZID=America/New_York:20240221T120000 LAST-MODIFIED:20240206T182839Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Flexible Reasoning with Large Language Models as Informal Logic Pr ograms UID:CAL-8a0292fd-8d13410f-018d-7fade941-00002bb4demobedework@mysite.edu URL:https://cs.duke.edu/events/flexible-reasoning-large-language-models-in formal-logic-programs X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-SPEAKER:Hongyuan Mei\, TTIC X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=bdhingra@cs.duke.edu:Bhu wan Dhingra X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Hongyuan Mei\, TTIC X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/hongyuanmeidukecalendar_20240206062839PM.j pg X-BEDEWORK-THUMB-IMAGE:/public/Images/hongyuanmeidukecalendar_202402060628 39PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Engineering CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240207T191030Z DESCRIPTION:LUNCH \nLunch will be served at 11:45 AM \n\nABSTRACT \nComput er networks are at the foundation of modern society. When they fail\, ban ks go offline\, airplanes stop flying\, emergency numbers stop working\, and businesses lose millions of dollars. Over the last decade\, the netwo rking community has made significant progress toward technology that can provide formal correctness guarantees for network behavior. I co-develop ed one such tool\, called Batfish\, that is in use at many large enterpri ses. I'll share key lessons from its evolution from a research prototype to an industrial-strength product\, such as how Datalog-based logic prog ramming had significant limitations and how binary decision diagrams (BDD s) proved highly versatile. I will also outline key challenges in using c urrent verification technology and what is needed to further improve netw ork reliability by an order of magnitude. \n\nSPEAKER BIO \nRatul Mahajan is a faculty member at the University of Washington (Paul G. Allen Schoo l of Computer Science). He is also the co-director of UW FOCI (Future of Cloud Infrastructure) and an Amazon Scholar. Ratul is a computer systems researcher with a networking focus and has worked on a broad set of topic s\, including network verification\, connected homes\, network programmin g\, optical networks\, Internet routing and measurements\, and mobile sys tems. Many of the technologies that he has helped develop are part of rea l-world systems at Microsoft and other companies. Ratul has been recogniz ed as an ACM Distinguished Scientist\, an ACM SIGCOMM Rising Star\, and a Microsoft Research Graduate Fellow. His papers have won the ACM SIGCOMM Test-of-Time Award\, the IEEE William R. Bennett Prize\, the ACM SIGCOMM Best Paper Awards (twice)\, and the HVC Best Paper Award. DURATION:PT1H DTSTAMP:20240207T191030Z DTSTART;TZID=America/New_York:20240223T120000 LAST-MODIFIED:20240207T191030Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Network Verification: Lessons Learned and Outlook UID:CAL-8a0292fd-8d13410f-018d-84fa92a4-000056dademobedework@mysite.edu URL:https://cs.duke.edu/events/network-verification-lessons-learned-and-ou tlook X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Engineering:/user/public-use r/Topics/Engineering X-BEDEWORK-SPEAKER:Ratul Mahajan\, UW-CSE X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=danyang@cs.duke.edu:Dany ang Zhuo X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Ratul Mahajan\, UW-CSE X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/ratulmahajandukecalendar_20240207071030PM. jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/ratulmahajandukecalendar_20240207071 030PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240213T201459Z DESCRIPTION:LUNCH \nLunch will be served at 11:45 AM. \n\nABSTRACT \nAs ma ssive language models (LMs) like GPT-4 dominate natural language processi ng and AI\, extreme-scale has become a clear and frequent theme for succe ss. However\, increasing model size is inherently at odds with the intere sts of a diverse user base and community of open researchers. The largest models are typically closed to the public\, extremely energy-intensive\, and difficult to conduct systematic and reproducible research on. \n\nIn this talk\, I will discuss my vision for effective natural language proc essing beyond scale alone. I will begin by describing alternative approac hes\, more efficient methods that work with compact language models to un lock hidden capabilities. I begin with inference-time algorithms that wor k on top of existing models to allow new functionality. Next\, I describe a method for distilling valuable knowledge from extreme-scale models int o compact LMs. Finally\, I will explain my work towards understanding the limits that even extreme-scale language models suffer from\, with a part icular focus on how such models differ from human intuitions. \n\nSPEAKER BIO \nPeter West is PhD student in the Paul G. Allen School of Computer Science & Engineering at the University of Washington\, working with Yeji n Choi. His research is focused on natural language processing and langua ge models\, particularly studying the capabilities and limits of both com pact and extreme-scale models. His work has received multiple awards\, in cluding best methods paper at NAACL 2022\, and outstanding paper awards a t ACL and EMNLP in 2023. His work has been supported in part by an NSERC PGS-D fellowship. Previously\, Peter received a BSc in computer science f rom the University of British Columbia. DURATION:PT1H DTSTAMP:20240213T201459Z DTSTART;TZID=America/New_York:20240304T120000 LAST-MODIFIED:20240213T201459Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Discovering Hidden Capabilities and Limits in Large Language Model s UID:CAL-8a0292fd-8d13410f-018d-a41bc60d-000075e1demobedework@mysite.edu URL:https://cs.duke.edu/events/discovering-hidden-capabilities-and-limits- large-language-models X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-SPEAKER:Peter West\, UW-CSE X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=tomasi@duke.edu:Carlo To masi X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Peter West\, UW-CSE X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/peterwestdukecalendar_20240213081459PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/peterwestdukecalendar_20240213081459 PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Engineering CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240216T180036Z DESCRIPTION:SNACKS: \nSnacks will be served at 3:45 PM. \n\nABSTRACT: \nAn increasing number of security and privacy researchers are conducting res earch with the intention of informing public policy discussions. I will t alk about how I got interested in public policy and discuss a wide range of research projects I've worked on over the past 25 years that had some impact on public policy discussions including research on the usability o f privacy tools\, the cost of reading privacy policies\, various types of privacy labels\, password policy\, and the California privacy choice ico n. \n\nSPEAKER BIO: \nLorrie Faith Cranor is the Director and Bosch Disti nguished Professor of the CyLab Security and Privacy Institute and FORE S ystems University Professor of Computer Science and of Engineering and Pu blic Policy at Carnegie Mellon University. She directs the CyLab Usable P rivacy and Security Laboratory (CUPS) and co-directs the Privacy Engineer ing master's program. In 2016 she served as Chief Technologist at the US Federal Trade Commission. She co-founded Wombat Security Technologies. Sh e is a fellow of the ACM\, IEEE\, and AAAS and a member of the ACM CHI Ac ademy. \n\nZOOM BROADCAST: \nDr. Cranor will be giving her lecture from N orth Carolina State University. As she is a TCSDLS speaker\, we will be p roviding direct access to her talk via Zoom broadcast in LSRC D106 at Duk e. \n\nTRIANGLE COMPUTER SCIENCE DISTINGUISHED LECTURER SERIES: \nThe com puter science departments at Duke University\, North Carolina State Unive rsity\, and the University of North Carolina at Chapel Hill joined forces to create the Triangle Computer Science Distinguished Lecturer Series. T he lecture series began in the 1995-1996 academic year\, and is made poss ible by grants from the U.S. Army Research Office\, rotated among the dep artments. DURATION:PT1H DTSTAMP:20240216T180741Z DTSTART;TZID=America/New_York:20240304T160000 LAST-MODIFIED:20240216T180741Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:From Password Requirements to IoT Cybersecurity Labels: Informing Public Policy with Research UID:CAL-8a0292fd-8d13410f-018d-b313d0af-0000056edemobedework@mysite.edu URL:https://cs.duke.edu/events/password-requirements-iot-cybersecurity-lab els-informing-public-policy-research X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Engineering:/user/public-use r/Topics/Engineering X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=michael.reiter@duke.edu: Mike Reiter X-BEDEWORK-SPEAKER:Lorrie Faith Cranor\, CMU X-BEDEWORK-DUKE-SERIES:Triangle Computer Science Distinguished Lecturer Se ries X-BEDEWORK-DUKE-WEBCAST:https://ncsu.zoom.us/j/97778793085?pwd=OG5CUW9GZHd Gelk0Rkk4Rmxab2lYQT09 X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Lorrie Faith Cranor\, CMU X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/dukecalendar4mar2024tcsdls_20240216060036P M.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/dukecalendar4mar2024tcsdls_202402160 60036PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240219T164336Z DESCRIPTION:LUNCH: \nLunch will be served at 11:45 AM \n\nABSTRACT: \nGene rative flow models learn a (possibly stochastic) mapping between source a nd target distributions. Common paradigms include diffusion models\, scor e matching models\, and continuous normalizing flows. In this talk I will first present methods for improved training of flow models using flow ma tching objectives using ideas from optimal transport. I will then show ho w these improved methods can be applied to the tasks of (1) modelling cel l dynamics\, which allow us to better understand disease programs - leadi ng to a new potential therapeutic pathway for triple-negative breast canc er and (2) generative protein design\, with applications to biologic drug discovery. \n\nSPEAKER BIO: \nAlex Tong is a postdoctoral researcher at Mila and Université de Montréal\, where he works with Yoshua Bengio at th e intersection of generative machine learning and biology with focuses on applications to cells and proteins. This work is part of a joint effort with Fabian Theis through the Helmholtz International Lab. He is also cof ounder of Dreamfold\, a Mila startup which builds generative models for p rotein design. DURATION:PT1H DTSTAMP:20240219T164336Z DTSTART;TZID=America/New_York:20240306T120000 LAST-MODIFIED:20240219T164336Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Flow Models with Applications to Cell Trajectories and Protein Des ign UID:CAL-8a0292fd-8d13410f-018d-c2406610-00005dafdemobedework@mysite.edu URL:https://cs.duke.edu/events/flow-models-applications-cell-trajectories- and-protein-design X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-SPEAKER:Alex Tong\, MILA and Dreamfold.ai X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=amink@cs.duke.edu:Alexan der Hartemink X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Alex Tong\, MILA and Dreamfold.ai X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/alextongdukecalendar_20240219044336PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/alextongdukecalendar_20240219044336P M-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Ethics CATEGORIES:Engineering CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240220T170723Z DESCRIPTION:LUNCH: \nLunch will be served at 11:45 AM. \n\nABSTRACT: \nEth ics discussions abound but translating "do no harm" into our work is frus trating at best\, and obfuscatory at worst. We can agree that keeping hum ans safe and in control is important\, but implementing ethics is intimid ating work. Learn how to wield your preferred technology ethics code to m ake systems that are accountable\, de-risked\, respectful\, secure\, hone st and usable. The presenter will introduce the topic of ethics and then step through a framework to guide teams successfully through this process . \n\nThis talk is for teams working on (or anticipating working on) emer ging technologies such as artificially intelligent (AI) systems. Attendee s do not need any previous experience or knowledge about ethics. \n\nSPEA KER BIO: \nCarol Smith is the AI Division Trust Lab Lead and a Principal Research Scientist at the Carnegie Mellon University (CMU)\, Software Eng ineering Institute. She leads research focused on development practices t hat result in trustworthy\, human-centered\, and responsible AI systems. Ms. Smith has been conducting research to improve the human experience wi th complex systems across industries for over 20 years. Since 2015 she ha s led research to integrate ethics and improve human experiences with AI systems\, autonomous vehicles\, and other complex and emerging technologi es. Ms. Smith is recognized globally as a leading researcher and user exp erience advocate and has presented over 250 talks and workshops in over 4 0 cities around the world. Her writing can be found in publications from organizations including AAAI\, ACM\, and the UXPA\, and she has taught co urses and lectured at CMU and other leading institutions. Ms. Smith is cu rrently an ACM Distinguished Speaker and a Working Group member of the IE EE P7008™ Standard. Ms. Smith holds a Master of Science degree in Human-C omputer Interaction from DePaul University. DURATION:PT1H DTSTAMP:20240220T170723Z DTSTART;TZID=America/New_York:20240321T120000 LAST-MODIFIED:20240220T170723Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Implementing Ethics in Emerging Technologies UID:CAL-8a0292fd-8d13410f-018d-c77c8724-000021f8demobedework@mysite.edu URL:https://cs.duke.edu/events/implementing-ethics-emerging-technologies X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Engineering:/user/public-use r/Topics/Engineering X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Ethics:/user/public-user/Top ics/Ethics X-BEDEWORK-SPEAKER:Carol Smith\, CMU X-BEDEWORK-DUKE-SERIES:ACM-W Distinguished Speaker X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=lisa@cs.duke.edu:Lisa Wu Wills X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Carol Smith\, CMU X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/carolsmithdukecalendar_20240220050723PM.jp g X-BEDEWORK-THUMB-IMAGE:/public/Images/carolsmithdukecalendar_2024022005072 3PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Technology CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20240221T165112Z DESCRIPTION:LUNCH: \nLunch will be served at 11:45 AM. \n\nABSTRACT: \nThe last decade saw a dramatic shift in how NLP systems are developed and de ployed: from bespoke models trained for specific tasks to a single Large Language Model (LLM) capable of solving a variety of tasks via in-context learning (ICL). In this talk\, I will discuss the problem of adapting LL Ms to out-of-distribution tasks and datasets. \n\nSPEAKER BIO: \nSilvio A mir is an assistant professor at the Khoury College of Computer Science a t Northeastern University. He completed his PhD at the University of Lisb on and was a postdoc at Johns Hopkins University before joining Northeast ern. Silvio Amir works on Natural Language Processing and Machine Learnin g methods to analyze personal and user generated text\, such as social me dia and clinical notes from Electronic Health Records. He is primarily in terested in tasks involving subjective\, personalized or user-level infer ences (e.g.\, opinion mining and digital phenotyping). In particular\, hi s work aims to improve the reliability\, interpretability and fairness of predictive models and analytics derived from these data. DURATION:PT1H DTSTAMP:20240221T165112Z DTSTART;TZID=America/New_York:20240408T120000 LAST-MODIFIED:20240221T165112Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Can One Model Rule Them All? Tailoring Large Language Models to Sp ecialized Domains\, Specific Populations\, and Unique Individuals UID:CAL-8a0292fd-8d13410f-018d-cc94125f-000043bfdemobedework@mysite.edu URL:https://cs.duke.edu/events/can-one-model-rule-them-all-tailoring-large -language-models-specialized-domains-specific X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Technology:/user/public-user /Topics/Technology X-BEDEWORK-SPEAKER:Silvio Amir\, Northeastern Univ. X-BEDEWORK-DUKE-SERIES:Duke Computer Science Colloquium X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=bdhingra@cs.duke.edu:Bhu wan Dhingra X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:0 X-BEDEWORK-IMAGE-X2:530 X-BEDEWORK-IMAGE-Y2:353.3333333333333 X-BEDEWORK-IMAGE-CROP-WIDTH:530 X-BEDEWORK-IMAGE-CROP-HEIGHT:353.3333333333333 X-BEDEWORK-IMAGE-ALT-TEXT:Silvio Amir\, Northeastern Univ. X-BEDEWORK-SUBMITTEDBY:cr390 for Computer Science (agrp_ArtsandSciences_Co mputerScience) X-BEDEWORK-IMAGE:/public/Images/silvioamirdukecalendar_20240221045112PM.jp g X-BEDEWORK-THUMB-IMAGE:/public/Images/silvioamirdukecalendar_2024022104511 2PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Lecture/Talk CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Main CONTACT;X-BEDEWORK-UID=18832edc-1b31c518-011b-40a23061-00000095:Hester\, G lenda CREATED:20241001T183248Z DESCRIPTION:While there have been huge advances in Machine Learning in rec ent years\, many of the successes have relied on immense amounts of train ing data. Especially for sequential-decision-making tasks (the realm of reinforcement learning)\, obtaining such data from online experience can take a very long time. On the other hand\, learning can often be dramati cally accelerated by leveraging human input\, for example as demonstratio ns of successful task executions\, as interventions to correct mistakes\, or simply as evaluative feedback separating "correct" actions from incor rect actions. This talk focuses on such Human-in-the-Loop Machine Learni ng for robotics tasks\, covering both navigations\, especially in tightly constrained spaces\, and manipulation in open-world settings. DURATION:PT1H DTSTAMP:20241015T153342Z DTSTART;TZID=America/New_York:20241028T120000 LAST-MODIFIED:20241015T153342Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Human-in-the-Loop Machine Learning for Robot Navigation and Manipu lation UID:CAL-8a00048d-91324965-0192-495b3be4-0000089fdemobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Lecture_Talk:/user/public-us er/Lectures_Conferences/Lecture_Talk X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-CS;X-BEDEWORK-PARAM-DESCRIPTION="/principals/users/agrp_PrattSc hool_BME,/principals/users/agrp_PrattSchool_CEE,/principals/users/agrp_Ar tsandSciences_ComputerScience,/principals/users/agrp_Institutes_Materials ScienceandEngineering,/principals/users/agrp_PrattSchool_ECE,":Biomedical Engineering (BME)\,Civil and Environmental Engineering (CEE)\,Computer S cience\,Duke Materials Initiative\,Electrical and Computer Engineering (E CE) X-BEDEWORK-SPEAKER:Peter Stone X-BEDEWORK-SUBMITTEDBY:ghester for Mechanical Engineering and Materials Sc ience (MEMS) (agrp_PrattSchool_MEMS) END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260114T151846Z DESCRIPTION:Addressing global security challenges requires enhanced radiat ion detection capabilities to meet the needs of organizations such as the IAEA\, CTBTO\, and the U.S. Government. Key priorities include the accur ate determination of radionuclide concentrations in environmental samples with improved speed\, sensitivity\, and adaptability to diverse sample p rofiles\, such as aged or minute samples. Advances in rare-event physics technologies\, originally developed for dark matter searches and neutrino experiments\, are proving critical in achieving the ultra-sensitive thre sholds needed for these missions. This presentation will explore the appl ication of next-generation groundwater age-dating methodologies as a case study for leveraging rare-event physics technologies. Additionally\, I w ill introduce the multidisciplinary Facility for Underground Science and Engineering (FUSE) under development at Savannah River National Laborator y\, highlighting its unique underground laboratory environment optimized for minimizing background radiation and enabling cutting-edge detection r esearch. DURATION:PT1H DTSTAMP:20260114T153211Z DTSTART;TZID=America/New_York:20260122T150000 LAST-MODIFIED:20260114T153211Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Advancing Ultra-Sensitive Radiation Detection: Applications of Rar e Event Physics to Global Security UID:CAL-8a00eca5-9af98aae-019b-bd167f74-00001d33demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-CS;X-BEDEWORK-PARAM-DESCRIPTION="/principals/users/agrp__Artsan dSciences_Physics,":Physics X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=sean.finch@duke.edu:Sean Finch X-BEDEWORK-SPEAKER:Henning Back X-BEDEWORK-DUKE-WEBCAST:https://duke.zoom.us/j/94471581751?pwd=yi0iSLM7JKg LehUQsWkmuayNEi2xjI.1 X-BEDEWORK-SUBMITTEDBY:asc for TUNL Seminar Series (agrp_TUNL_SeminarSerie s) X-BEDEWORK-IMAGE-ALT-TEXT:Photo of TUNL Building X-BEDEWORK-IMAGE-X1:5 X-BEDEWORK-IMAGE-Y1:36 X-BEDEWORK-IMAGE-X2:2084 X-BEDEWORK-IMAGE-Y2:1422 X-BEDEWORK-IMAGE-CROP-WIDTH:2079 X-BEDEWORK-IMAGE-CROP-HEIGHT:1386 X-BEDEWORK-IMAGE:/public/Images/TUNL_0_20260114033211PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/TUNL_0_20260114033211PM-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260126T155841Z DESCRIPTION:In the late 1950s the U.S. Atomic Energy Commission (AEC) plan ned Project Plowshare\, a series of tests designed to show the peaceful u ses for nuclear explosions. One plan was to create an artificial harbor i n Cape Thompson\, Alaska using multiple hydrogen bombs in Project Chariot . However\, nearby\, the Iñupiat residents and local scientists fought to keep their land pristine. Taking on Edward Teller and the US Government\ , this group of Alaskan citizens thwarted the project and\, in the proces s\, started a movement. This talk will discuss the project\, the results of the fight\, and the far-reaching implications of this project. DURATION:PT1H DTSTAMP:20260126T155841Z DTSTART;TZID=America/New_York:20260212T150000 LAST-MODIFIED:20260126T155841Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CONFIRMED SUMMARY:Could nuclear physics help us understand Alzheimer’s? UID:CAL-8a00eca5-9af98aae-019b-fb075bc1-000016bademobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-CS;X-BEDEWORK-PARAM-DESCRIPTION="/principals/users/agrp__Artsan dSciences_Physics,":Physics X-BEDEWORK-SPEAKER:Shelly R. Lesher X-BEDEWORK-DUKE-WEBCAST: https://duke.zoom.us/j/94471581751?pwd=yi0iSLM7JK gLehUQsWkmuayNEi2xjI.1 X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=sean.finch@duke.edu:Sean Finch X-BEDEWORK-IMAGE-X1:72 X-BEDEWORK-IMAGE-Y1:86 X-BEDEWORK-IMAGE-X2:967.5 X-BEDEWORK-IMAGE-Y2:683 X-BEDEWORK-IMAGE-CROP-WIDTH:895.5 X-BEDEWORK-IMAGE-CROP-HEIGHT:597 X-BEDEWORK-IMAGE-ALT-TEXT:Prof. Shelley Lesher X-BEDEWORK-SUBMITTEDBY:asc for TUNL Seminar Series (agrp_TUNL_SeminarSerie s) X-BEDEWORK-IMAGE:/public/Images/LesherPhoto-1024x683_20260126035841PM.jpg X-BEDEWORK-THUMB-IMAGE:/public/Images/LesherPhoto-1024x683_20260126035841P M-thumb.png END:VEVENT BEGIN:VEVENT CATEGORIES:Natural Sciences CATEGORIES:Lectures/Conferences CATEGORIES:Utilities CATEGORIES:Panel/Seminar/Colloquium CATEGORIES:Main CONTACT;X-BEDEWORK-UID=00f1fcdb-0f068baf-010f-068baf83-00000004:None CREATED:20260414T160240Z DESCRIPTION:The advent of next-generation facilities capable of delivering high-intensity\, high-resolution gamma-ray beams and ultra-short\, high- repetition laser pulses is opening new avenues in nuclear astrophysics. N arrow\, tunable gamma-ray beams (ranging from a few keV to tens of MeV) o ffer unprecedented opportunities to measure extremely small cross section s (on the order of μb or less) relevant to key astrophysical nuclear reac tions. These precise measurements are crucial for determining astrophysic al S-factors\, which underpin our understanding of stellar evolution.\n\n In recent years\, the ASFIN group\, together with international collabora tors\, has developed advanced detector arrays optimized for the measureme nt of excitation functions and angular distributions. These systems\, inc luding LHASA and ELISSA\, are ideally suited for nuclear astrophysics stu dies induced by gamma-ray beams. In this contribution\, I will present th e recent developments and discuss their application to ongoing and future experimental campaigns. DURATION:PT1H DTSTAMP:20260415T192758Z DTSTART;TZID=America/New_York:20260416T150000 LAST-MODIFIED:20260415T192758Z LOCATION;X-BEDEWORK-UID=18832edc-1bffd55f-011c-2ea4458d-000003df:LSRC D106 STATUS:CANCELLED SUMMARY:Nuclear Astrophysics with Gamma Beams UID:CAL-8a003384-9cc4d1fe-019d-8cbb0a98-00007157demobedework@mysite.edu X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Main:/user/public-user/Utili ties/Main X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Panel_Seminar_Colloquium:/us er/public-user/Lectures_Conferences/Panel_Seminar_Colloquium X-BEDEWORK-ALIAS;X-BEDEWORK-PARAM-DISPLAYNAME=Natural Sciences:/user/publi c-user/Topics/Natural Sciences X-BEDEWORK-CS;X-BEDEWORK-PARAM-DESCRIPTION="/principals/users/agrp__Artsan dSciences_Physics,":Physics X-BEDEWORK-STUDENT-CONTACT;X-BEDEWORK-PARAM-EMAIL=sean.finch@duke.edu:Sean Finch X-BEDEWORK-SPEAKER:Giovanni Luca Guardo X-BEDEWORK-DUKE-WEBCAST:https://duke.zoom.us/j/94471581751?pwd=yi0iSLM7JKg LehUQsWkmuayNEi2xjI.1 X-BEDEWORK-IMAGE-X1:0 X-BEDEWORK-IMAGE-Y1:43 X-BEDEWORK-IMAGE-X2:2088 X-BEDEWORK-IMAGE-Y2:1435 X-BEDEWORK-IMAGE-CROP-WIDTH:2088 X-BEDEWORK-IMAGE-CROP-HEIGHT:1392 X-BEDEWORK-IMAGE-ALT-TEXT:Photo of TUNL Building X-BEDEWORK-SUBMITTEDBY:asc for TUNL Seminar Series (agrp_TUNL_SeminarSerie s) X-BEDEWORK-IMAGE:/public/Images/TUNL_0_20260414040241PM.png X-BEDEWORK-THUMB-IMAGE:/public/Images/TUNL_0_20260414040241PM-thumb.png END:VEVENT END:VCALENDAR