BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Automatically Discovering Learning Algorithms with Hardware Constr
 aints
DTSTART:20241204T111500
DTEND:20241204T121500
DTSTAMP:20260501T224348Z
UID:b9a3ffd9640d76ce13a8187183a0f4cc88cca8347c3fc94dcc404380
CATEGORIES:Conferences - Seminars
DESCRIPTION:  Esteban Real (Google DeepMind)\nCan learning emerge from a 
 search process in-silico? Our AutoML-Zero work at Google DeepMind shows th
 at a simple evolutionary search process can automatically discover modern 
 learning techniques from scratch. Without knowledge of machine learning th
 e method discovers algorithms by combining simple primitives such as addit
 ions and multiplications into a functional learning algorithm. In this typ
 e of evolutionary search learning emerges naturally when algorithm surviva
 l depends on its performance on multiple tasks. These discovery methods tr
 ansfer to realistic setups where we can find novel algorithms for ML optim
 ization and robot adaptation. Importantly we can meaningfully shape the pr
 operties of the discovered algorithm by constraining the environment on wh
 ich the algorithm evolves. I will speculate that using hardware abstractio
 ns as such a constraint is a promising direction for finding new paradigms
  of neural computation.
LOCATION:GC C3 30 https://plan.epfl.ch/?room==GC%20C3%2030
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR