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:20260420T160446Z
DTSTART:20260429T120000Z
DTEND:20260429T130000Z
SUMMARY:SQUIDS Seminar: Optimizing the Dynamics of the Frequency Bias in 
 Fourier Features Neural Networks
UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}m220-mo7dyc
 1e-xtzdib
DESCRIPTION:Speaker: Professor Matias Courdurier (Pontifical Catholic Uni
 versity of Chile)\n\nAbstract: \nImplicit Neural Representations often l
 earn signals capturing low-frequency structure before high-frequency det
 ails\, a phenomenon known as Frequency Bias. Fourier Feature networks ca
 n reduce this effect by encoding inputs with sinusoidal components sampl
 ed from a prescribed distribution.\n\nIn this talk we will present how\,
  in the Neural Tangent Kernel setting\, it is possible to derive a PDE t
 hat describes the evolution\, during training\, of different frequencies
  and use the exact form of the PDE to select initialization distribution
 s to tune or suppress this frequency bias. We formulate an explicit opti
 mization procedure to pick an optimal first-layer parameter distribution
 \, including a Gaussian-restricted version\, yielding faster early-stage
  convergence. Through experiments we also validate the theorical predict
 ion and\, by testing across tasks and deeper FF architectures\, we show 
 robustness of the proposed approach beyond the simplified setting in whi
 ch the PDE is obtained.\n\nThis is a joint work with Juan Jose Molina\, 
 Mircea Petrache and Francisco Sahli Costabal.\n
STATUS:TENTATIVE
TRANSP:TRANSPARENT
CLASS:PUBLIC
LOCATION:Frank Adams Room 2\, Alan Turing Building\, Manchester
END:VEVENT
END:VCALENDAR