BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260421T090513Z
LOCATION:Bldg. 6 - Room 102
DTSTART;TZID=Europe/Stockholm:20260629T140000
DTEND;TZID=Europe/Stockholm:20260629T143000
UID:submissions.pasc-conference.org_PASC26_sess106_msa132@linklings.com
SUMMARY:Towards Green AI: Reducing the Environmental Footprint of LLM Work
 loads
DESCRIPTION:Shashikant Ilager (University of Amsterdam) and Ivona Brandic 
 (TU Wien)\n\nContemporary large-scale computing systems are becoming incre
 asingly heterogeneous, complex, and decentralized, driven by growing digit
 ization demands. This transformation is most evident in the rapid rise of 
 Generative AI applications, which have fueled the expansion of hyperscale 
 data centers, now often referred to as AI factories. While these infrastru
 ctures enable unprecedented computational capabilities, they impose high e
 nvironmental costs through substantial energy consumption and carbon emiss
 ions.\nAssessing and managing resource utilization in such distributed sys
 tems presents complex challenges, requiring consideration of hardware hete
 rogeneity, diverse power characteristics, varied user workloads, and poten
 tial performance degradation from energy-saving interventions.\nThis talk 
 presents our recent research on data-driven modeling and optimization tech
 niques for energy-aware management of AI applications. First, I discuss ru
 ntime techniques to enhance the energy efficiency of LLM-based code assist
 ants. Second, I detail adaptive quantization of KV cache in LLMs to reduce
  memory and compute requirements. Through these contributions, this talk d
 emonstrates how LLM-based applications can be optimized through hardware-s
 oftware co-design strategies that balance computational performance with s
 ustainability objectives, offering practical pathways toward environmental
 ly responsible AI deployment.\n\nDomain: Engineering, Computational Method
 s and Applied Mathematics\n\nSession Chairs: Denisa-Andreea Constantinescu
  (EPFL), Florina Ciorba (University of Basel), and Can Hankendi (Boston Un
 iversity)\n\n
END:VEVENT
END:VCALENDAR