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:20231101T093405Z
DTSTART:20231115T143000Z
DTEND:20231115T160000Z
SUMMARY:SQUIDS Seminar - How mathematics can inspire novel deep learning
architectures: two case studies
UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}s27r-lofk9e
1e-qzt3gw
DESCRIPTION:Schedule:\n- 2:30pm: pretalk\n- 3:05pm: research talk 'How ma
thematics can inspire novel deep learning architectures: two case studie
s'\n \nAbstract: Deep learning is revolutionising the modern world\, inc
luding applied mathematics. Yet\, it has infamously been described as "a
lchemy"\, and one place that alchemy can arise is in the choice of archi
tecture. Can we leverage mathematical insights to design deep learning a
rchitectures in a more principled way? \n \nIn this talk\, I will give t
wo examples of this from my own research. In the first\, I will talk abo
ut my recent preprint with Carola-Bibiane SchĂ¶nlieb\, Subhadip Mukherjee
\, and Zakhar Shumaylov on learned regularisation for inverse problems i
n imaging\; how we were inspired by (non)convex analysis to design a nov
el input weakly convex neural network\, and used this to learn an advers
arial convex-nonconvex regulariser with provable guarantees and which ov
ercame the numerical issues of previous adversarial regularisers. In the
second\, I will talk about ongoing research with Martin Benning\, Jonas
Latz\, Lisa Kreusser\, and James Rowbottom\, on designing a deep learni
ng architecture for image segmentation based on the graph Merriman-Bence
-Osher segmentation method. This method is particularly promising becaus
e the graph this method learns on the inputted image provides some built
-in explainability of the outputted segmentation.\n
STATUS:TENTATIVE
TRANSP:TRANSPARENT
CLASS:PUBLIC
LOCATION:G.114\, Alan Turing Building\, Manchester
END:VEVENT
END:VCALENDAR