BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Date iCal//NONSGML kigkonsult.se iCalcreator 2.20.2//
METHOD:PUBLISH
X-WR-CALNAME;VALUE=TEXT:Eventi DIAG
BEGIN:VTIMEZONE
TZID:Europe/Paris
BEGIN:STANDARD
DTSTART:20191027T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RDATE:20201025T030000
TZNAME:CET
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20200329T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.19963.field_data.0@www.diag.uniroma1.it
DTSTAMP:20260601T031551Z
CREATED:20200518T154048Z
DESCRIPTION:Fabrizio Silvestri è risultato vincitore della procedura selett
 iva ad 1 posto di Professore di I fascia presso il Dipartimento di Ingegne
 ria Informatica\, Automatica e Gestionale Antonio Ruberti\, Settore Concor
 suale 09/H1\, SSD ING-INF/05 - codice concorso 2019POE007\, bandito con De
 creto Rettorale  2757/2019 del 19/09/2019\, i cui atti sono stati approvat
 i con  Decreto Rettorale n. 1272/2020 del 13/05/2020. Nell'ambito della pr
 ocedura ai fini della chiamata da parte del Consiglio di dipartimento\, Fa
 brizio Silvestri terrà un seminario ed una lezione pubblica sulle attività
  di ricerca da lui svolte e in corso di svolgimento. Il seminario e la lez
 ione  saranno svolte in modalità telematica Venerdì 22 Maggio\, alle ore 1
 4:00.Per partecipare al seminario e alla lezione\, connettersi all’indiriz
 zo https://meet.google.com/jvg-snvt-qmeIl titolo della lezione verrà estra
 tto il giorno 21 Maggio 2020 alle ore 13:45 su meet.google.com/cxf-pxjk-tq
 d in presenza della Direttrice prof.ssa Tiziana Catarci. Title: Robustness
  in Language Models: From Misspelling Resistant Embedding to Fact Checkabl
 e Text Generation.AbstractIn this talk\, We will review two recent results
  in the field of Neural Language Modelling. In the first part I will show 
 a technique to improve n-gram embeddings\, namely FastText\, and making th
 em more robust to misspellings. Traditional word embeddings are good at so
 lving lots of natural language processing (NLP) downstream problems such a
 s documentation classification and named-entity recognition (NER). However
 \, one of the drawbacks is a lack of capability on handling out-of-vocabul
 ary (OOV). Misspelling Oblivious (word) Embedding (MOE) overcomes this lim
 itation by using a combination of self-supervised and supervised language 
 modelling.The second part will show a mechanism to generate and assess the
  fact-checkability of neural generated text. We argue that verifiability\,
  i.e.\, the consistency of the generated text with factual knowledge\, is 
 a suitable metric for measuring this cost. We use an automatic fact-checki
 ng system to calculate new metrics as a function of the number of supporte
 d claims per sentence and find that sampling-based generation strategies\,
  such as top-k\, indeed lead to less verifiable text. Based on this findin
 g\, we introduce a simple and effective generation strategy for producing 
 non-repetitive and more verifiable (in comparison to other methods) text.F
 inally\, we present some lines of research inspired by these recent findin
 gs.
DTSTART;TZID=Europe/Paris:20200522T140000
DTEND;TZID=Europe/Paris:20200522T140000
LAST-MODIFIED:20210425T094108Z
LOCATION:Meet:  https://meet.google.com/jvg-snvt-qme
SUMMARY:Robustness in Language Models: From Misspelling Resistant Embedding
  to Fact Checkable Text Generation (public) - \n\n\n  \n  \n\n    \n\n\nFa
 brizio\n\n\nSilvestri  \n\n  \n\n    \n\n\n\n\n\nProfessore ordinario\n\n
 \npagina personale\n\nstanza: \n\nB209\n\ntelefono: \n\n+39 0677274015\n\n
 Member of: \n\n  \n\n  \n\n    \n\nBiografia: \n\n\n\nFabrizio Silvestri i
 s a Full Professor and the coordinator of the Ph.D. in Data Science\, at D
 ipartimento di Ingegneria informatica\, automatica e gestionale (DIAG) of 
 the University of Rome\, La Sapienza. His research interests lie in Artifi
 cial Intelligence\, and in particular\, Fabrizio Silvestri deals with mach
 ine learning applied to web search problems and natural language processin
 g. He is the author of more than 150 papers in international journals and 
 conference proceedings. It holds nine industrial patents. He is the holder
  of the 'test-of-time' award at the ECIR 2018 conference for an article pu
 blished in 2007. He is the holder of three best paper awards and other int
 ernational awards. Fabrizio Silvestri spent eight years abroad in industri
 al research laboratories (Yahoo! and Facebook). At Facebook AI\, Fabrizio 
 Silvestri has directed research groups to develop artificial intelligence 
 techniques to combat malicious actors who use the Facebook platform for ma
 licious purposes (hate speech\, misinformation\, terrorism\, etc.) Fabrizi
 o Silvestri has a Ph.D. in computer science awarded by the University of P
 isa with a thesis entitled: 'High-Performance Issues in Web Search Engines
 : Algorithms and Techniques.'\n\n\nqualifica_rr: \n\nProfessors\n\nRiconos
 cimenti e premi: \n\nTest of Time Award for the paper “Sorting Out the Doc
 ument Identifier Assignment Problem”. The paper was selected among those p
 ublished more than 10 years before and according to the nomination the pap
 er represents a mileston in Inverted Index compression. Nowadays\, no pape
 r on inverted index compression is published without firstly applying the 
 technique presented in that paper.\nNominated for the Super Star Award at 
 Yahoo. Fewer than 10% of Yahoos are nominated for Super Star each year. Su
 per Stars are Yahoos who go above and beyond expectations\, contribute to 
 products that make Yahoo the world’s indispensable guide to digital inform
 ation\, and exude positivity that brings out the best in others.\nYahoo Te
 ch Pulse 2014 Best Paper Award for the paper What App Are You Going to Ope
 n Next? Personalized Mobile App Prediction for Aviate.\nYahoo Patent Miles
 tone Award in recognition of my Yahoo! Patent Inventions.\nRecipient of th
 e Yahoo Labs Excellence Award Program (LEAP) for my work with the Advertis
 ing science team that has been included in production code.\nOne of the re
 cipients of the 2011 Yahoo! Faculty Research and Engagement Program (FREP)
  for the Computational Advertisement\, Machine Learning and Search & Web M
 ining Category (http://labs.yahoo.com/node/596)\nYoung Investigator Award 
 at ISTI-CNR. Most scientifically prolific young scholar in 2010.\nBest Pap
 er Award Runner Up at 4th ACM Conference on Web Search and Data Mining (WS
 DM 2011) with the paper “Identifying Task-based Sessions in Search Engine 
 Query Logs”. Co-authors: Claudio Lucchese\, Salvatore Orlando\, Raffaele P
 erego\, Gabriele Tolomei.\nBest Paper Award at 29th European Conference on
  Information Retrieval (ECIR 2007) with the paper “Sorting Out the Documen
 t Identifier Assignment Problem”.\nFinalist for the ERCIMs Cor Baayen Awar
 d: selected as one of the two representatives from Italy for year 2005.\nB
 est Paper Award at 2004 IEEE/WIC/ACM International Conference on Web Intel
 li- gence (WI 2004) with the paper “An Online Recommender System for Large
  Web Sites”. Co-author: Ranieri Baraglia.
URL;TYPE=URI:https://www.diag.uniroma1.it/node/19963
END:VEVENT
END:VCALENDAR