BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:AI for Meta Science
DTSTART:20261123T080000Z
DTEND:20261127T160000Z
DTSTAMP:20261029T062800Z
UID:indico-event-11640@indico.ijclab.in2p3.fr
CONTACT:david.rousseau@ijclab.in2p3.fr
DESCRIPTION:Overview\nHackathons are announced\, registration has reopened
 \nArtificial Intelligence (AI) has gained significant attention for its ap
 plications in science. This workshop will focus on its potential use beyon
 d primary scientific data analysis pipelines\, hence the "meta" in AI for 
 meta-science.\nThis workshop is mainly funded by AISSAI\, CNRS's interdisc
 iplinary centre for AI\, and is hosted by Institut Pascal\, Université Pa
 ris-Saclay.\nThe workshop is structured to favour discussions. Each theme 
 will be covered in half a day to a full day\, featuring 45-minute keynote 
 talks and a series of short contributed talks\, with ample time for discus
 sion\, which is well-suited to Institut Pascal. Each participant will have
  a desk in shared offices with access to meeting rooms (and\, of course\, 
 free coffee machines) and will be provided with lunches\, a workshop dinne
 r on Thursday\, and additional social events.\nTwo hackathons will be run 
 in parallel for two half-days\, Mad Scientist\, and From Literature to Ben
 chmark (see below).\nAttendance is limited to 70 scientists\, by invitatio
 n or upon application till 6th Nov (menu on the left)\; applications will
  be evaluated based on applicants’ expertise in at least one of the topi
 cs covered\, as well as the goal of ensuring a diverse group of participan
 ts. Young researchers including PhD students are especially encouraged to 
 apply.\nConfirmed speakers\nConfirmed speakers include so far:\n\nDavid Ch
 amont (IJCLab) Labobots school\n\nAdnan Ghribi (GANIL) Summary of accelera
 tor data hackathon\n\n\nDamien Gratadour (CNRS) AI control of telescopes\n
 \n\nVerena Kain (CERN) AI control of accelerators\n\nFrancois Lanusse (CEA
 ) LightCone : agentic AI assisted science\nPierre Manil (CEA) AI and mecha
 nical design\n\nPietro Vischia (Oviedo) AI assisted experiment co design\n
 \n\nMad Scientist hackathon is organised by Savannah Thais (Hunter\, NY) a
 nd Nathan Suri (Yale)From Literature to benchmark hackathon is organised b
 y Joao Goncalves (U Bonn) and Romina Raoofian (U Bonn)\n\nThemes\nSeveral 
 key areas have been identified: \n\nAI for managing the operation of very
  large scientific apparatuses\, such as particle accelerators\, telescopes
  or large gravitational wave detectors. Can these facilities be made fully
  autonomous in terms of control and monitoring?  \nAI for optimising larg
 e computing resource infrastructures by intelligently pre-loading data and
  scheduling data processing tasks to maximise throughput.\nAI for enhancin
 g experiment design\, including the use of fully differentiable simulators
  to optimise experiment sensitivity and its integration with generative de
 sign techniques in engineering. Do experiments to be analysed by AI algori
 thms need to be designed differently?\nLarge Language Models (LLMs) for ma
 naging\, processing\, and making available the internal documentation asso
 ciated with large-scale scientific experiments\, including\, but not limit
 ed to\, the software codebase.\nAI for epistemology: can an LLM not only e
 xtract the most up-to-date information but also analyse the dynamics of sc
 ientific thought by tracking the emergence and decline of ideas through th
 e study of scientific peer-review papers\, as well as drafts and discussio
 n documents?\nAI for scientific discovery: to what extent can AI be used t
 o become a major player in the scientific process\, by making hypotheses\,
  connecting insights and designing experiments\nLast but not least: what d
 oes it mean to do a PhD in such an environment? How should future scientis
 ts be trained?  \n\n \nHackathons\n.\nMad Scientist\nMad Scientist\, the
  first red-teaming effort built for AI agents for scientific research. As 
 a hackathon\, participants will be tasked with designing "attacks" to deli
 berately misalign an honest agentic pipeline created to search for new par
 ticles at the Large Hadron Collider. Attacks can be in the form of input p
 oisoning\, agent misalignment\, or even multi-agent collusion. The aim of 
 the exercise is to use our knowledge as science practitioners to intuit wh
 ere an agent could fail in a way that would pass standard control structur
 es. These attacks can then be used to build more robust and safe scientifi
 c agents as well as inform scientists who collaborate with these agents on
  critical failure modes to watch out for.\nFrom Literature to Benchmark\nA
  new major goal of th AI and physics community is to develop Foundation Ph
 ysics Model. These models trained on very large multimodal datasets would 
 be capable of performing tasks like particle identification\, inference or
  generation. The purpose of this hackathon is to define automatic system c
 apable to parse the litterature and identify the most relevant benchmarks 
 for foundation models.\nContact: David Rousseau\, IJCLab\, rousseau@ijclab
 .in2p3.fr\n\nhttps://indico.ijclab.in2p3.fr/event/11640/
IMAGE;VALUE=URI:https://indico.ijclab.in2p3.fr/event/11640/logo-127194463.
 png
LOCATION:Institut Pascal
URL:https://indico.ijclab.in2p3.fr/event/11640/
END:VEVENT
END:VCALENDAR