BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//https://techplay.jp//JP
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALDESC:Towards Understanding Deep Learning through the Lens of Statis
 tical Physics
X-WR-CALNAME:Towards Understanding Deep Learning through the Lens of Statis
 tical Physics
X-WR-TIMEZONE:Asia/Tokyo
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
BEGIN:STANDARD
DTSTART:19700101T000000
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:986783@techplay.jp
SUMMARY:Towards Understanding Deep Learning through the Lens of Statistical
  Physics
DTSTART;TZID=Asia/Tokyo:20250929T140000
DTEND;TZID=Asia/Tokyo:20250929T150000
DTSTAMP:20260528T014701Z
CREATED:20250924T060149Z
DESCRIPTION:イベント詳細はこちら\nhttps://techplay.jp/event/98678
 3?utm_medium=referral&utm_source=ics&utm_campaign=ics\n\nThis is an onlin
 e seminar. Registration is required.\n\n【Date】Monday\, September 29th
 \, 14:00 - 15:00\n【Speaker】Wei Huang\, AIP Deep Learning Theory Team\
 n\nTITLE: \nTowards Understanding Deep Learning through the Lens of Stati
 stical Physics\n\nABSTRACT:\nLarge foundation models such as GPT and stab
 le diffusion models are transforming science and society. Despite their i
 mpressive capabilities\, we still lack clear principles to explain why th
 ey generalize\, when they fail\, and how emergent abilities arise. To add
 ress these questions\, I approach deep learning from the perspective of s
 tatistical physics\, viewing neural networks as complex systems with many
  interacting components. This allows us to apply tools such as mean-field
  theory and phase transitions to reveal hidden laws of learning. In this 
 talk\, I will first introduce kernel methods as a mean-field description 
 of infinite-width networks\, showing how neural tangent kernels help expl
 ain the role of orthogonal initialization and how graph neural tangent ke
 rnels clarify the trainability of graph neural networks. I will then pres
 ent a framework of feature learning theory\, where signal–noise models 
 serve as a foundational model for deep learning\, analogous to the Ising 
 model in statistical physics. This framework explains phenomena such as b
 enign overfitting in Transformers and provides insights into feature lear
 ning in diffusion models. Together\, these studies show how statistical-p
 hysics-inspired approaches can uncover hidden principles of deep learning
  and foundation models.
LOCATION:オンライン
URL:https://techplay.jp/event/986783?utm_medium=referral&utm_source=ics&utm
 _campaign=ics
END:VEVENT
END:VCALENDAR