BEGIN:VCALENDAR
PRODID:-//AddEvent Inc//AddEvent.com v1.7//EN
VERSION:2.0
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
BEGIN:STANDARD
DTSTART:20261101T010000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20260308T030000
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
DESCRIPTION:Short Bio: Hasan is a PhD student at King Abdullah Univesity of Science and Technology (KAUST) under the supervision of Bernard Ghanem. Hasan has worked in multi topics such as robustness\, 3D understanding\, LLM agents\, and continual learning. The work to be presented in this session is on data diversity and self-supervised training which was completed during Hasan's internship at The University of Oxford.\n\nAbstract: We explore the impact of training with more diverse datasets\, characterized by the number of unique samples\, on the performance of self-supervised learning (SSL) under a fixed computational budget. Our findings consistently demonstrate that increasing pretraining data diversity enhances SSL performance\, albeit only when the distribution distance to the downstream data is minimal. Notably\, even with an exceptionally large pretraining data diversity achieved through methods like web crawling or diffusion-generated data\, among other ways\, the distribution shift remains a challenge. Our experiments are comprehensive with seven SSL methods using large-scale datasets such as ImageNet and YFCC100M amounting to over 200 GPU days. Code and trained models will be available at https://github.com/hammoudhasan/DiversitySSL\n\n------\n\nCreate your own Add to Calendar links with addevent.com/r/a \n
X-ALT-DESC;FMTTYPE=text/html:Short Bio: Hasan is a PhD student at King Abdullah Univesity of Science and Technology (KAUST) under the supervision of Bernard Ghanem. Hasan has worked in multi topics such as robustness, 3D understanding, LLM agents, and continual learning. The work to be presented in this session is on data diversity and self-supervised training which was completed during Hasan's internship at The University of Oxford.<br><br>Abstract: We explore the impact of training with more diverse datasets, characterized by the number of unique samples, on the performance of self-supervised learning (SSL) under a fixed computational budget. Our findings consistently demonstrate that increasing pretraining data diversity enhances SSL performance, albeit only when the distribution distance to the downstream data is minimal. Notably, even with an exceptionally large pretraining data diversity achieved through methods like web crawling or diffusion-generated data, among other ways, the distribution shift remains a challenge. Our experiments are comprehensive with seven SSL methods using large-scale datasets such as ImageNet and YFCC100M amounting to over 200 GPU days. Code and trained models will be available at <a href="https://github.com/hammoudhasan/DiversitySSL" target="_blank">https://github.com/hammoudhasan/DiversitySSL</a><br /><br />------<br /><br />Create your own Add to Calendar links with addevent.com/r/a <br>
UID:eac557aa48e841528f70ab653c80de9daddeventcom
SUMMARY:[C4AI] Hasan Abed Al Kader Hammoud - On Pretraining Data Diversity for Self-Supervised Learning (Geo Asia)
DTSTART;TZID=America/Los_Angeles:20240513T090000
DTEND;TZID=America/Los_Angeles:20240513T100000
DTSTAMP:20260922T105716Z
TRANSP:OPAQUE
STATUS:CONFIRMED
SEQUENCE:0
LOCATION:https://meet.google.com/yhv-tiir-ava
X-MICROSOFT-CDO-BUSYSTATUS:BUSY
BEGIN:VALARM
TRIGGER:-PT30M
ACTION:DISPLAY
DESCRIPTION:Reminder
END:VALARM
END:VEVENT
END:VCALENDAR