BEGIN:VCALENDAR
METHOD:PUBLISH
PRODID:Microsoft Exchange Server 2010
VERSION:2.0
X-WR-CALNAME:Calendar
BEGIN:VTIMEZONE
TZID:W. Europe Standard Time
BEGIN:STANDARD
DTSTART:16010101T030000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RRULE:FREQ=YEARLY;INTERVAL=1;BYDAY=-1SU;BYMONTH=10
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:16010101T020000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
RRULE:FREQ=YEARLY;INTERVAL=1;BYDAY=-1SU;BYMONTH=3
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
DESCRIPTION:Title: Why are inverse folding models good zero-shot predictors
  of protein thermodynamic stability?\n\n\n\nJes Frellsen<https://frellsen.
 org/>\, Technical University of Denmark\, Cognitive Systems\n\n\n\nAbstrac
 t: Inverse folding models are trained to recover sequences from structures
 \, yet they have emerged as highly effective zero-shot predictors of prote
 in stability. How can we understand this connection? In this talk\, I unpa
 ck the theoretical assumptions that connect the amino acid preferences of 
 an inverse folding model to the free-energy considerations that govern the
 rmodynamic stability. Drawing on concepts from probability theory and stat
 istical physics\, I will show that commonly used heuristics can be interpr
 eted as simplistic approximations and that more principled alternatives em
 pirically yield considerable performance gains.\n\nBio: Jes Frellsen is an
  Associate Professor at the Technical University of Denmark (DTU). He rece
 ived his PhD in Bioinformatics from the University of Copenhagen. Before j
 oining DTU\, he was a postdoc in the Machine Learning Group at the Univers
 ity of Cambridge and an Associate Professor at the IT University of Copenh
 agen. At DTU\, he leads a research group on probabilistic machine learning
  and generative AI. His methodological contributions include work on missi
 ng data\, uncertainty quantification\, and out-of-distribution detection\,
  with applications in bioinformatics\, physics\, recommender systems\, and
  remote sensing. He has authored more than 70 research articles and book c
 hapters\, with numerous at the premier machine learning venues. He contrib
 utes actively to the community through conference chairing and as a foundi
 ng co-organiser of the GeMSS summer school on deep generative models. He w
 as recently awarded the Jorck’s Foundation Research Prize for his contri
 butions to methods for handling incomplete data and enabling the safe and 
 reliable use of AI.\n\n
UID:040000008200E00074C5B7101A82E00800000000A6C0E4D75E64DC01000000000000000
 0100000000ABAA81EE816CA47A5B4154BD434B2F0
SUMMARY:ML seminar - Jes Frellsen - Why are inverse folding models good zer
 o-shot predictors of protein thermodynamic stability?
DTSTART;TZID=W. Europe Standard Time:20260311T133000
DTEND;TZID=W. Europe Standard Time:20260311T150000
CLASS:PUBLIC
PRIORITY:5
DTSTAMP:20260606T234910Z
TRANSP:OPAQUE
STATUS:CONFIRMED
SEQUENCE:1
LOCATION:IDA Alan Turing [40] (Campus Valla)
X-MICROSOFT-CDO-APPT-SEQUENCE:1
X-MICROSOFT-CDO-BUSYSTATUS:BUSY
X-MICROSOFT-CDO-INTENDEDSTATUS:BUSY
X-MICROSOFT-CDO-ALLDAYEVENT:FALSE
X-MICROSOFT-CDO-IMPORTANCE:1
X-MICROSOFT-CDO-INSTTYPE:0
X-MICROSOFT-DONOTFORWARDMEETING:FALSE
X-MICROSOFT-DISALLOW-COUNTER:FALSE
X-MICROSOFT-REQUESTEDATTENDANCEMODE:DEFAULT
X-MICROSOFT-ISRESPONSEREQUESTED:FALSE
END:VEVENT
BEGIN:VEVENT
DESCRIPTION:Title: Learning causally sound and interpretable composite endp
 oints for clinical trials\n\nSpeaker: Fredrik D. Johansson<https://www.fre
 djo.com/>\, Chalmers University of Technology\, Healthy AI Lab\n\nAbstract
 : Randomized clinical trials are considered the gold standard evidence for
  learning about the causal effects of medical interventions\, but have nat
 ural limitations on scope and length. This often rules out targeting long-
 term outcomes of interest\, such as mortality or cardiovascular disease\, 
 as these endpoints won’t be observed for most participants during the le
 ngth of the trial. Instead\, researchers turn to surrogate endpoints that 
 are associated with the primary outcome of interest and can be observed du
 ring the trial. This presents a problem: What constitutes a good surrogate
 ? In theory\, a good surrogate is one for which the effect of the treatmen
 t is predictive of its effect on the primary outcome\, but the definition 
 alone does not reveal how to find such a variable. More than that\, to be 
 useful in a clinical trial\, the surrogate must be approved by a regulator
 y body when registering the trial\, necessitating its interpretability. In
  this talk\, I will discuss the implications of this\, algorithms that can
  provably learn composite surrogates from observational data\, and situati
 ons where there is no hope to find a good surrogate.\n\nLocation: Alan Tur
 ing\n
UID:040000008200E00074C5B7101A82E00800000000C3A1E4E25E64DC01000000000000000
 010000000DBE93C4037366C4F9866EA488629AB2C
SUMMARY:ML seminar - Fredrik D. Johansson - Learning causally sound and int
 erpretable composite endpoints for clinical trials
DTSTART;TZID=W. Europe Standard Time:20260408T133000
DTEND;TZID=W. Europe Standard Time:20260408T150000
CLASS:PUBLIC
PRIORITY:5
DTSTAMP:20260606T234910Z
TRANSP:OPAQUE
STATUS:CONFIRMED
SEQUENCE:1
LOCATION:IDA Alan Turing [40] (Campus Valla)
X-MICROSOFT-CDO-APPT-SEQUENCE:1
X-MICROSOFT-CDO-BUSYSTATUS:BUSY
X-MICROSOFT-CDO-INTENDEDSTATUS:BUSY
X-MICROSOFT-CDO-ALLDAYEVENT:FALSE
X-MICROSOFT-CDO-IMPORTANCE:1
X-MICROSOFT-CDO-INSTTYPE:0
X-MICROSOFT-DONOTFORWARDMEETING:FALSE
X-MICROSOFT-DISALLOW-COUNTER:FALSE
X-MICROSOFT-REQUESTEDATTENDANCEMODE:DEFAULT
X-MICROSOFT-ISRESPONSEREQUESTED:FALSE
END:VEVENT
BEGIN:VEVENT
DESCRIPTION:Title (tentative): Probabilistic Machine Learning Meets Large-s
 cale Deep Learning Models\n\nSpeaker: Martin Trapp<https://trappmartin.git
 hub.io/>\, KTH\n\nAbstract (tentative):\nModern deep learning models\, suc
 h as LLMs and agentic systems\, suffer from overconfidence and lack well-c
 alibrated epistemic uncertainties. However\, they are on such an enormous 
 scale that classical Bayesian methods either do not apply directly or are 
 challenging to scale. In this talk\, I will discuss how probabilistic mach
 ine learning methods\, particularly Bayesian methods\, provide a valuable 
 and urgently needed toolset for large-scale deep learning models\, but req
 uire careful treatment to obtain effective and efficient approaches that d
 eliver meaningful improvements. I will draw on recent work to discuss how 
 stochastic tokenisation improves the robustness of LLMs\, how Bayesian fra
 meworks can be meaningfully integrated into agentic AI systems\, and how l
 inks to tractable probabilistic models enable us to quantify quantisation 
 errors in deep learning models in a principled way.\n\nLocation: Alan Turi
 ng\n
UID:040000008200E00074C5B7101A82E00800000000275CBFEA0EE1DC01000000000000000
 01000000058237D7830D4314F9A92ADDB8EA5B918
SUMMARY:ML seminar - Martin Trapp - Probabilistic Machine Learning Meets La
 rge-scale Deep Learning Models
DTSTART;TZID=W. Europe Standard Time:20260610T133000
DTEND;TZID=W. Europe Standard Time:20260610T150000
CLASS:PUBLIC
PRIORITY:5
DTSTAMP:20260606T234910Z
TRANSP:OPAQUE
STATUS:CONFIRMED
SEQUENCE:2
LOCATION:IDA Alan Turing [40] (Campus Valla)
X-MICROSOFT-CDO-APPT-SEQUENCE:2
X-MICROSOFT-CDO-BUSYSTATUS:BUSY
X-MICROSOFT-CDO-INTENDEDSTATUS:BUSY
X-MICROSOFT-CDO-ALLDAYEVENT:FALSE
X-MICROSOFT-CDO-IMPORTANCE:1
X-MICROSOFT-CDO-INSTTYPE:0
X-MICROSOFT-DONOTFORWARDMEETING:FALSE
X-MICROSOFT-DISALLOW-COUNTER:FALSE
X-MICROSOFT-REQUESTEDATTENDANCEMODE:DEFAULT
X-MICROSOFT-ISRESPONSEREQUESTED:FALSE
END:VEVENT
END:VCALENDAR