BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Emanuele Rodolà (Sapienza University) DTSTART:20220713T073000Z DTEND:20220713T083000Z DTSTAMP:20260423T010620Z UID:GaML/9 DESCRIPTION:Title: Fro m Sound to Metric Priors: A New Paradigm for Shape Generation\nby Eman uele Rodolà (Sapienza University) as part of Workshop on Geometry and Mac hine Learning\n\n\nAbstract\nSpectral and metric geometry are at the heart of various problems in computer vision\, graphics\, pattern recognition\, and machine learning. Ultimately\, the core reason for their success can be traced down to questions of stability and to the informativeness of the eigenvalues of certain operators. In this talk\, I will discuss and show tangible examples of such properties and showcase some dramatic implicatio ns on a selection of notoriously hard problems in computer vision and grap hics. First\, I will address the question of whether one can recover the s hape of a geometric object from its vibration frequencies (‘hear the sha pe of the drum’)\; while theoretically the answer to this question is ne gative\, little is known about the practical possibility of using the spec trum for shape reconstruction and optimization. I will introduce a numeric al procedure called isospectralization\, as well as a data-driven variant\ , showing how this *practical* problem is solvable. Then\, I will discuss the increasingly popular task of designing an effective generative model f or deformable 3D shapes. I will demonstrate how injecting metric distortio n priors into a simple geometric reconstruction loss can lead to the forma tion of a very informative latent space\, which can be trained with extrem ely scarce data (less than 10 examples) and still yield competitive genera tion quality as well as aiding geometric disentanglement.\n LOCATION:https://researchseminars.org/talk/GaML/9/ END:VEVENT END:VCALENDAR