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:20251026T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20250330T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.29379.field_data.0@www.ugov-ricerca.uniroma1.it
DTSTAMP:20260409T205300Z
CREATED:20250623T200459Z
DESCRIPTION:Abstract: The paradigm shift of Large Language Models (LLMs) is
  in part due to bringing down the time-to-content creation by orders of ma
 gnitude. While we can be generally content with using LLMs to carry out si
 mple and risk-free tasks\, at their core LLMs remain autoregressive neural
  networks\, thus we cannot control their “hallucinations” nor we can quant
 ify when distribution-shift happens. This makes it very hard for software 
 engineers to use LLMs as a practical and reliable tool in their everyday w
 ork. Also\, training LLMs remains an insurmountable task for those with le
 ss resources\, which makes NLP research harder. This is interesting\, beca
 use there are no guarantees that the current sizes of LLMs are optimal. Th
 e costs of performing model selection are way too high to justify a proper
  exploration of the architectural space\, so there is a tangible possibili
 ty that the right size might be way smaller from what we are used to\, pot
 entially leading to a democratization of (L)LM research. In this talk\, we
  will address two separate topics. First\, we will briefly see how to quan
 titatively measure the “stability” of LLMs to prompt rewritings\, so that 
 software engineers can at least select and debug LLMs according to two new
 ly introduced metrics. Then\, we will spend more time on a new technique t
 o train neural networks so that their size adapts to the task within a sin
 gle training run\, up to an infinite size. We will show how the neural net
 work “orders information by its importance”\, opening a whole new set of c
 onsiderations when discussing neural networks training.    Bio: Federico E
 rrica is a Senior Research Scientist at NEC Laboratories Europe\, which he
  joined since 2022. He received his PhD in Computer Science from the Unive
 rsity of Pisa in 2022\, with a thesis on Bayesian graph machine learning s
 upervised by Davide Bacciu and Alessio Micheli. His current interests revo
 lve around the understanding deep graph networks’ behavior and the ability
  of neural networks to adapt their architecture during training. Federico 
 is part of the Intelligent Software Systems group\, which performs basic r
 esearch on computational science and AI agents topics.
DTSTART;TZID=Europe/Paris:20250703T140000
DTEND;TZID=Europe/Paris:20250703T140000
LAST-MODIFIED:20260326T110808Z
LOCATION:B203
SUMMARY:Present and Future of (L)LMs: considerations on reliability and eff
 iciency - Federico Errica
URL;TYPE=URI:http://www.ugov-ricerca.uniroma1.it/node/29379
END:VEVENT
END:VCALENDAR