Page 26 - BBC Knowledge - October 2017 IN

Science
Discoveries
T H E Y D I D W H A T ? !
MA TH S
COMPUTATIONAL

ORIGAMI TAKES

A BIG LEAP FORWARD

An MIT professor of computer science and the boundaries of the original sheet of paper,
an assistant professor in civil engineering at and minimises the number of seams. “It’s a totally
the University of Tokyo have joined forces to different strategy for thinking about how to make
come up with a better way of… making paper a polyhedron,” said Demaine.
rabbits. Or rather, they have created an algorithm If you’ve ever unfolded a paper cup from
that enables the creation of any 3D shape from the water cooler, and ended up with a circular
a single sheet of a given material. piece of paper, that’s the perfect example of how ROBOT
MIT’s Prof Erik Demaine has previous the new algorithm works – the outer edge of the TAUGHT TO
experience in this area: his 1999 PhD thesis circle ends up as the rim of the cup. Demaine’s
described the same thing. The difference, old method, however, would have created a non- COMPOSE
though, is that his previous algorithm essentially watertight cup shape by winding a thin strip of
involved taking a long, thin strip of paper or other paper into a coil. MUSIC
material and winding it into the desired shape. The technique could have practical
This tends to leave you with lots of seams in the applications in manufacturing, particularly WHAT DID THEY DO?
finished 3D shape, and is inefficient in terms of in areas such as designing and building Computer scientists at
the amount of paper (or other material) required. spacecraft, where materials efficiency is Georgia Institute of
The new algorithm, on the other hand, preserves of paramount importance. Technology in the US have
taught a robot to compose
its own musical pieces, and
The new origami then play them on the
algorithm can make
any shape from marimba – an instrument
a single sheet similar to a xylophone.
of material
HOW DID THEY DO THAT?
The robot – nicknamed
‘Shimon’ – was fed nearly
5,000 complete
compositions, ranging from
pop songs to classical
pieces, and over two
million smaller fragments
such as riffs, solos and
codas. Using deep learning
techniques, its AI system
then analysed the material
and devised its own set
of rules for composition.
Using these rules, it then
‘wrote’ and played
recognisably musical
creations of its own.

WHY DID THEY DO THAT?
Project leader Mason
Bretan is interested in
exploring the possibilities
of AI and computer
learning in music
composition. Maybe
the first robot masterpiece
26 is just around the corner.
OCTOBER 2017