IEEE Spectrum - North American - March 2015 - 13

Getty iMaGes

superIor sensors
The Atlas detector is one of four placed
a rou nd t he few poi nt s where t he
counter rotating protons collide headon. Benjamino Di Girolamo, the technical coordinator of the Atlas experiment,
says that workers had to replace a layer
of detectors, or "pixels," that were close
to the beam pipe because of radiation
damage. However, such a replacement
proved impossible because of structural constraints in the detector, so they
decided to build an insert, called the
IBL (insertable B layer) that would fit
between the beam pipe and the innermost layer of the detector assembly.
The IBL was then able to take advantage of better radiation-hardened detector pixels, which were developed by
a research program at CERN. But the
most important aspect of the IBL is
that it improves the precision of Atlas.
"Having a layer at the beginning of a
[particle] track is very important to us.
If you have high precision very near the
start of the track, then you can even
relax the precision you need in the
outer layers [of the detector assembly],"
says Di Girolamo.
He reports that the insertion of the
IBL, which was a very delicate operation, was successful. "The IBL, for its
operation, has been cooled down and
is now fully working," he says.

nEwS

neered new hardware itself-the CAMAC
(computer automated measurement and
control) industrial data-acquisition standard and capacitive touch screens are
examples. Although CERN now makes
fewer custom systems, Neufeld says
there is a need for specialized highspeed, radiation-hardened electronics that can operate in strong magnetic
fields. And a bigger upgrade is in the
works. For experiments scheduled to
start in 2019, the data flow is expected
to increase by two orders of magnitude.
"This will require a significant change,
and new technologies will have to be
put in," says Neufeld.

ais vs. Poker
making computers unbeatable at texas
hold 'em could lead to big breakthroughs
in artificial intelligence

Life is not a game. but there possible plays in even the simplest
are similarities. That's why poker game is a far greater chalit's worthwhile to invent artificial- lenge because each player has hidintelligence algorithms that can win den cards-information hidden from
games. One such AI has now finally the opponent. In that sense, poker
solved one of the simplest versions is an "imperfect-information game,"
of poker. It's a crucial first step on similar to real-world scenarios with
a potentially long road toward various degrees of uncertainty.
"The solutions for imperfectbeating human poker champions
in more complex versions of the information games require comgames. This isn't just about brag- puters to handle the additional
ging rights: Poker playing can train complication of not knowing exactly
computer algorithms to tackle the what the game's state is, such as not
complexities of real-world chal- knowing an opponent's hand," says
lenges in security and medicine, Neil Burch, a computer scientist at
where the available information is the University of Alberta, in Edmonton, Canada. "Such techniques
rarely perfect or complete.
The many possibilities for decep- require more computer memory
tion in poker means there are a huge and computing power."
Burch and his colleagues laid
number of possible plays, even in a
limited version with just two play- out their algorithm's solution to
ers. Computers have solved simpler "heads-up limit" Texas Hold 'em in
games such as Connect Four and the journal Science in January. In
checkers by figuring out the per- AI parlance, it's only a "weak" solufect, unbeatable strategy for each tion to a specific version of poker.
move starting from the beginning The game has just two players,
of each game. But analyzing all the fixed bet amounts, and a fixed

SPECTRUM.IEEE.ORG

nORTh aMERICan

MaR 2015

http://SPECTRUM.IEEE.ORG

Table of Contents for the Digital Edition of IEEE Spectrum - North American - March 2015