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:20260324T162137Z
DTSTART:20260513T100000Z
DTEND:20260513T110000Z
SUMMARY:AI-Fun & ELLIS Invited Speaker Series | Nikolay Malkin
UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}x1en-mkdxxx
 68-nm88s3
DESCRIPTION:For May's AI-Fun and ELLIS invited speaker series\, we will h
 ave Nikolay Malkin from the University of Edinburgh.\n\nTitle: Inferring
  stochastic dynamics without data: from diffusion samplers to discrete S
 chrödinger bridges\n\nAbstract: \nProbabilistic models that approximate 
 a distribution by transporting particles from a source distribution to t
 he target following a learnt dynamics model have seen rapid development 
 and adoption in recent years: indeed\, diffusion models and continuous n
 ormalising flows show success in generative modelling for various domain
 s. I will describe the less-known use of such dynamics-based models as v
 ariational families: fitting their parameters to sample distributions fr
 om which no samples are available but an unnormalised target density can
  be queried. This problem has many algorithmic faces\, with connections 
 to entropic reinforcement learning\, optimal transport\, stochastic cont
 rol\, and sequential Monte Carlo. Our recent work has extended algorithm
 s for diffusion sampling to the discrete-space case and to learning brid
 ge dynamics between two distributions without access to samples from bot
 h. Applications include sampling Boltzmann densities of molecular confor
 mations\, inverse problems and conditional generation under pretrained g
 enerative model priors\, and accelerating particle-based algorithms for 
 Bayesian inference (in the continuous case) and inference over probabili
 stic model structure (e.g.\, Bayesian program induction and symbolic reg
 ression) and alignment of discrete-latent image generative models (in th
 e discrete case).\n\nBio:\nNikolay Malkin is a Chancellor's Fellow in In
 formatics at the University of Edinburgh and a fellow of CIFAR's Learnin
 g in Machines and Brains programme. Their research focuses on algorithms
  for probabilistic inference and Bayesian machine learning\, with applic
 ations in generative modelling\, neurosymbolic AI\, and machine reasonin
 g. Within machine learning\, their work explores modelling of Bayesian p
 osteriors over high-dimensional and structured variables\, induction and
  discovery of compositional structure in generative models\, and uncerta
 inty-aware reasoning in language and formal systems. Their work has foun
 d applications in pure and applied sciences\, including inverse imaging\
 , remote sensing\, discovery of novel biological and chemical structures
 \, and\, most recently\, robot control. Dr Malkin holds a PhD in mathema
 tics from Yale University (2021) and was previously a postdoctoral resea
 rcher at Mila – Québec AI Institute in Montréal (2021 to 2024).\n\nIf yo
 u are unable to attend in person\, please follow the ticketsource link p
 rovided to register and then check the registration confirmation for the
  Teams link.
STATUS:TENTATIVE
TRANSP:TRANSPARENT
CLASS:PUBLIC
LOCATION:Lecture Theatre 1.3\, Kilburn Building\, Manchester
END:VEVENT
END:VCALENDAR