WHAT IS MACHINE LEARNING?
Machine learning is a type
of artificial intelligence that
focuses on enabling a computer
to learn new information all by
itself. Some learning methods
allow computers to find patterns
in large amounts of data, such
as identifying similar sets of
genes across a selection
of DNA sequences. Others can
cluster data into different groups,
allowing them to find different
patterns of fraud or normal
behaviour in credit card
transactions, for example.
Others are taught to recognise
data by viewing many different
examples, so they can
understand text or different
objects in a video. Still others
learn the ‘shape’ of data sothat they can predict what might
come next, enabling them to
anticipate where an object may
move, or how the stock market
might change.
Over many decades, all
these different learning methods
have grown from two main
sources of inspiration:
statistical mathematics and
biology. Most recently, some
of the biology-inspired methods
such as genetic algorithms
(based on natural evolution)
and deep learning (inspired
by the way that neurons learn
in the brain), combined with
some clever new maths, have
produced some of the most
impressive results we’ve seen
in robotics.the current capabilities of modern autopilots,”
explains Baomar.
The AI uses a neural network that operates
in a similar manner to the human brain,
with many different neural nodes arranged
in tiers and each one solving a different part
of the task simultaneously. Each successive tier
recieves the output from the previous tier rather
than the raw input. The nodes each have their
own bank of knowledge built up from their
original programming rules, plus anything
they’ve experienced.
Just as a human pilot may be simultaneously
using one part of their brain to move muscles,
one part to assess instruments and another to
speak, the AI uses many separate parts of its
brain to solve all the different problems of flying.
The AI learns directly from observing human
pilots, watching their every move in microscopic
detail in order to learn how to cope with whatever
gets thrown at it. It can then apply those skills .to
novel situations, flying new aircraft in scenarios
and conditions that it has never seen previously.
The system is designed to complement human
pilots rather than replace them, but Baomar hopes
the AI will improve air safety dramatically.
IRON CHEF
Researchers at the University of Maryland have
taken a similar observational approach and used it
L E F T: In the
last few years,
we’ve seen
cars that can
drive and park
themselves
FA R L E F T:
Patrick Tresset
with Paul,
the robot that
can draw
portraits