|
November 27, 2014
at 4:10 p.m.
MP102
McLennan Physical Laboratories (MP)
255 Huron Street
(campus
map)
|
Geoffrey Hinton, Computer Science at University of Toronto
Deep Learning
I will give a brief history of deep learning explaining what it is,
what kinds of task it should be good for and why it was
largely abandoned in the 1990's. I will then describe how
ideas from statistical physics were used to make deep learning
work much better. Finally I will describe how deep learning
is now used by Google for speech recognition and object
recognition and how it may soon be used for machine translation.
|
|
January 22, 2015
at 4:10 p.m.,
MP102
McLennan Physical Laboratories (MP)
255 Huron Street
(campus
map)
|
Karin Dahmen, University of Illinois at Urbana Champaign
Universal quake statistics: from nanopillars to earthquakes
The deformation of many solid materials is not continuous, but discrete,
with intermittent slips similar to earthquakes. Here, we suggest that
the statistical distributions of the slips, such as the slip-size
distributions, reflect tuned criticality, with approximately the same
regular (power-law) functions, and the same tunable exponential cutoffs,
for systems spanning 13 decades in length, from tens of nanometers
to hundreds of kilometers; for compressed nano-crystals, to amorphous
materials, to earthquakes. The similarities are explained by a simple
analytic model, which suggests that results are transferable across
scales. This study provides a unified understanding of fundamental
properties of shear-induced deformation in systems ranging from nanocrystals
to earthquakes. It also provides many new predictions for future experiments
and simulations. The studies draw on methods from the theory of phase
transitions, the renormalization group, and numerical simulations.
Connections to other systems with avalanches, such as magnets and
neuron firing avalanches in the brain are also discussed.
|
