Physics

The general purpose computer can run any program we can express in
symbolic logic; that makes it the go-to tool for accomplishing any task
that can be reduced to a computable function, and that's why software is
eating the world and cars and colliders and airplanes and pacemakers and
toasters are all just turning into computers in fancy cases.

That also means that every policy problem we can imagine will eventually
involve a computer -- and thus involve a lawmaker insisting that it must
be possible to make a computer that can run every program except for
some program that creates a problem in the world.

We don't know how to make that computer, but we can approximate it by
creating a computer with some spyware-like program running on it that
checks to see whether you're running a "naughty" program and shuts it down.

That's a spider we swallow to catch a fly. Here's the bird we swallow to
catch the spider: laws like Canada's C-11 make it a crime to
investigate, circumvent, or point out defects in systems that have these
spyware-like processes.

Companies have noticed that this offers some real upsides for them: they
can use these spyware-like processes to prevent users their customers
from using their own consumables (e.g. printer ink), parts, service
depots, apps -- anything that the company can command a high margin on,
provided that it's illegal for a customer to choose someone else's products.

What a deadly combination: companies are rapidly expanding the
constellation of devices that are locked in this way, and once a device
is locked in this way, it can't be investigated, its defects can't be
disclosed, and it can't be modified to improve it, mitigate its flaws,
and protects its owner from cybersecurity risk.

Peer review exists out of recognition that there is no way to know if
you're right until you let your enemies try to prove you wrong; in
creating a monopolistic right to control the use of products after they
were sold, Parliament also created the right for companies to enjoin the
most fundamental task of scientific knowledge creation: finding mistakes
and pointing them out.

This is urgent: our world is made of computers, those computers are
designed to treat their owners as their enemies, and security
researchers can't audit those computers without risking civil and
criminal reprisals. That's a catastrophe in the making.

Can computers think? They can certainly calculate - with staggering speed and ever-increasing power - and they have driven scientific and technological advances that would have been impossible without them. Even so, we would like to believe that, for some puzzles, there's no substitute for old-fashioned human intuition. But this view may be changing.

A new breed of machine learning algorithms have begun knocking down cognitive milestones that, until recently, scientists believed were still decades away. Major advances are being made in computer vision, language translation, autonomous robotic action, and other complex applications. At the same time, these new algorithms are helping scientists accelerate discovery in physics.

This stunning progress poses as many questions as answers: What are the fundamental possibilities and limits of machine learning? Can we create true human-level artificial intelligence, and how might its thoughts differ from our own? What new breeds of computer will fuel artificial intelligence and, conversely, how will artificial intelligence enable new forms of computing?

In his public lecture at Perimeter Institute, Roger Melko will explore how computers have helped humanity solve increasingly complex puzzles, and ask which challenges, if any, only human intuition is equipped to tackle.

The surprising success of learning with deep neural networks poses two fundamental challenges: understanding why these networks work so well and what this success tells us about the nature of intelligence and our biological brain. Our recent Information Theory of Deep Learning shows that large deep networks achieve the optimal tradeoff between training size and accuracy, and that this optimality is achieved through the noise in the learning process.

In this talk, I will focus on the statistical physics aspects of our theory and the interaction between the stochastic dynamics of the training algorithm (Stochastic Gradient Descent) and the phase structure of the Information Bottleneck problem. Specifically, I will describe the connections between the phase transition and the final location and representation of the hidden layers, and the role of these phase transitions in determining the weights of the network.

Based partly on joint works with Ravid Shwartz-Ziv, Noga Zaslavsky, and Shlomi Agmon.

Two years ago the Laser Interferometer Gravitational-wave Observatory.

(LIGO) announced the first direct detection of gravitational waves; minute distortions in space-time caused by cataclysmic events far away in the universe.  Very recently, the merger of a binary neutron star system was detected by both of the Advanced LIGO detectors and the Advanced Virgo detector in Italy, triggering a massive follow-up campaign by ground and space-based telescopes.  A counterpart to the gravitational-wave source was located, and transient emission was detected from gamma rays to radio.  I will talk about the sources of the signals we detected, the physics behind the detectors, and prospects for the future of this emerging field.