Prof. Masayuki Ohzeki
(Tohoku University, Tokyo Institute of Technology, Japan)
Title: Quantum Annealing and Its Application to Real World
Quantum annealing is a generic solver of combinatorial
optimization problem and is implemented by a hardware known as
the D-Wave quantum annealer.
In this talk, we introduce future directions of its
application, while showing several practical applications,
namely the control of the automated guided vehicles in factory
and evacuation system from disaster after big earthquake etc.
Prof. Ali Sheikholeslami
(University of Toronto, Canada)
Digital Annealer: A Stochastic Search for Global Optimum
As Moore's law nears the end of its time, the search for
continued improvement in performance has focused on the
architecture-level and system-level innovations.
At the system level, we resort to stochastic moves to solve
hard optimization problems, where the goal is to minimize
(or maximize) a function of many variables (in the order of
1000's) in a fraction of a second.
These optimization problems are predominant in engineering,
health, finance, and environment.
In engineering, for example, we often wish to allocate
resources to tasks, or schedule tasks given limited resources,
to minimize waste.
In health, we wish to maximize radiation to a tumor in a
patient's body while sparing the healthy organs surrounding
the tumor. This indeed requires optimization of the density
of an X-ray beam as it rotates 360 degrees around the patient.
In this keynote speech, we will walk you through a stochastic
journey of a Markov Chain Monte Carlo (MCMC) process where we
try to find the global minimum of a quadratic function of 1024
We will demonstrate how employing several techniques in
hardware parallelism including parallel tempering (deploying
several parallel hardware blocks exploring the solution space
at various "temperatures" and occasionally exchanging their
states), parallel trial, and parallel update, can provide
significant speedup, allowing CMOS to live far beyond the end
of CMOS scaling.
Dr. Alan Mishchenko
(University of California Berkeley, USA)
Title: Boolean Logic Networks for Machine Learning
This talk explores the use of logic networks and Boolean
methods in machine learning (ML). Both synthesis and
verification are addressed. On the synthesis side, we
introduce a novel ML model based on logic networks, which
can potentially replace neural networks in some applications.
The advantages are, substantially reduced evaluation latency,
simpler hardware implementation (no need for memories and
arithmetic operations), straight-forward design automation.
The challenge is, matching the accuracy of neural networks.
On the verification side, we use logic networks, in particular,
and-inverter graphs (AIGs), to detect overfitting in any ML
model using only the AIG representation of the model and the
training data, assuming that the evaluation data is not
available or cannot be trusted.