BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Critical time-dependent phenomena in diffusion generative models
DTSTART;VALUE=DATE-TIME:20260414T214000Z
DTEND;VALUE=DATE-TIME:20260414T214700Z
DTSTAMP;VALUE=DATE-TIME:20260408T063410Z
UID:indico-contribution-325@fisindico.uniandes.edu.co
DESCRIPTION:Speakers: William Montaño (Egresado)\nIn recent years\, diffu
 sion generative models have become state-of-the-art for tasks such as imag
 e\, video\, and audio generation\, among others. More recently\, there has
  been growing interest in studying the statistical mechanics of these mode
 ls\, driven by the observation of apparent phase transitions during the sa
 mpling process. More specifically\, a symmetry breaking that resembles the
  one encountered in the Ising model. In this proposal\, a theoretical desc
 ription of the diffusion models is presented\, explaining what diffusion m
 odels are\, how they can be studied from perspective of equilibrium statis
 tical mechanics\, and how critical phenomena emerge in a simple case. Addi
 tionally\, a simple simulation using a feed forward network and two delta 
 functions as the initial distributions provides some insight into the mode
 l's behavior near the critical point. The main objectives include a deeper
  investigation of this critical phenomenon\, with particular focus on ques
 tions concerning the relationship between data dimensionality and the numb
 er of spins in the Ising model\, as well as the emergence of scaling-free 
 properties. Furthermore\, the development of new models is proposed to all
 ow for a more detailed observation and analysis of these critical behavior
 s.\n\nhttps://fisindico.uniandes.edu.co/event/23/contributions/325/
LOCATION:Universidad Nacional Edificio 564
URL:https://fisindico.uniandes.edu.co/event/23/contributions/325/
END:VEVENT
END:VCALENDAR