BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Pedro Domingos (University of Washington)
DTSTART:20240215T170000Z
DTEND:20240215T180000Z
DTSTAMP:20260423T003247Z
UID:MPML/112
DESCRIPTION:Title: <a href="https://researchseminars.org/talk/MPML/112/">D
 eep Networks Are Kernel Machines</a>\nby Pedro Domingos (University of Was
 hington) as part of Mathematics\, Physics and Machine Learning (IST\, Lisb
 on)\n\n\nAbstract\nDeep learning's successes are often attributed to its a
 bility to automatically discover new representations of the data\, rather 
 than relying on handcrafted features like other learning methods. In this 
 talk\, however\, I will show that deep networks learned by the standard gr
 adient descent algorithm are in fact mathematically approximately equivale
 nt to kernel machines\, a learning method that simply memorizes the data a
 nd uses it directly for prediction via a similarity function (the kernel).
  This greatly enhances the interpretability of deep network weights\, by e
 lucidating that they are effectively a superposition of the training examp
 les. The network architecture incorporates knowledge of the target functio
 n into the kernel. The talk will include a discussion of both the main ide
 as behind this result and some of its more startling consequences for deep
  learning\, kernel machines\, and machine learning at large.\n
LOCATION:https://researchseminars.org/talk/MPML/112/
END:VEVENT
END:VCALENDAR