The role of scale in deep neural networks

Ethan Dyer and Guy Gur-Ari (Google)

17-May-2021, 19:30-20:30 (3 years ago)

Abstract: Modern deep learning has made remarkable progress across a wide variety of domains. A key ingredient to this success has been increasingly big models trained on increasingly large datasets. I will discuss theoretical and empirical work attempting to understand how performance improves with scale and distinguishing problems which can be solved by scale alone from those where new ideas are needed.

other condensed mattersoft condensed matterstatistical mechanicsstrongly correlated electronssuperconductivitygeneral relativity and quantum cosmologyHEP - theorymathematical physicschaotic dynamicsfluid dynamicsquantum physics

Audience: researchers in the topic


Kadanoff seminars

Organizers: Luca Delacretaz*, Nima Afkhami-Jeddi
*contact for this listing

Export talk to