BEGIN:VCALENDAR
PRODID:-//Columba Systems Ltd//NONSGML CPNG/SpringViewer/ICal Output/3.3-
 M3//EN
VERSION:2.0
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20251124T125540Z
DTSTART:20251124T140000Z
DTEND:20251124T150000Z
SUMMARY:Jason Hartford -- Maths in the Life Sciences Seminar [ONLINE]
UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}urf-mfuzx2x
 2-2bhyze
DESCRIPTION:[**Note\, this seminar will now take place online only**]\n\n
 Join us for this seminar by Jason Hartford (Manchester) as part of the M
 aths in the Life Sciences seminar series (and the online North West Semi
 nar Series in Mathematical Biology and Data Sciences in collaboration wi
 th Liverpool Universities). \n\nTitle: Scaling Biological Representation
  Learning and Experimental Design\n\nAbstract: High-throughput cellular 
 assays now generate data at a scale that enables highly predictive pheno
 typic models\, yet our ability to extract mechanistic insight from these
  systems remains limited. I will discuss our recent work on scaling mask
 ed-autoencoders on microscopy images and show that the resulting represe
 ntations are surprisingly effective at recovering known biological struc
 ture. I will then describe our work on mechanistic interpretability tech
 niques for decomposing these embeddings into interpretable\, biologicall
 y aligned features. Finally\, I will discuss experimental design in the 
 context of these large models. I will formalize a hierarchy of interacti
 on models—ranging from scalar viability readouts to full distributional 
 predictions—to characterize when single-perturbation information is suff
 icient to generalize to combinations. This perspective leads to an effic
 ient experimental-design strategy: treat non-interaction as the default\
 , use additive composition to generate predictions\, and focus measureme
 nts on identifying failures of this assumption.  \n\nThe talk will be st
 reamed via Teams\, please contact carl.whitfield@manchester.ac.uk or igo
 r.chernyavsky@manchester.ac.uk for the zoom link\, or sign up to the mai
 ling list.\n\nTo subscribe to the mailing list for this event series\, p
 lease send an e-mail with the phrase “subscribe math-lifesci-seminar” in
  the message body to listserv@listserv.manchester.ac.uk
STATUS:TENTATIVE
TRANSP:TRANSPARENT
CLASS:PUBLIC
LOCATION:Online
END:VEVENT
END:VCALENDAR