12 | New Scientist | 1 August 2020
CosmologyARTIFICIAL intelligence has been
trained to identify individual birds.
The system is being developed for
biologists studying wild animals,
but could be adapted to help people
recognise birds in their surroundings.
André Ferreira at the Center for
Functional and Evolutionary Ecology
in Montpellier, France, started the
project while studying how sociableweavers contribute to their colonies.
This is normally done by putting
coloured tags on their legs and
sitting by nests to watch them,
which is very time-consuming.
So Ferreira and his colleagues
turned to AI. The difficult part is
getting the photographs required
to train the system. “We need
thousands of pictures of the
same individual,” says Ferreira.
The researchers solved this
problem by putting RFID tags
on the birds, which triggered
cameras at bird feeders.The system has so far been tested
on captive zebra finches, wild great
tits and wild sociable weavers. Tests
with photographs that weren’t used
for training reveal its accuracy is
around 90 per cent for a single
image (Methods in Ecology and
Evolution, doi.org/d438).
For now, the system is still quite
limited. It has only been trained onpictures of the back of birds, as that
is the view biologists usually get
when observing behaviour. It might
also fail if the appearance of a bird
changes, such as during moulting.
However, Ferreira thinks that
all these issues can be overcome
if given large-enough data sets.
He and his team are now setting
up cameras to take pictures from
multiple angles, not just the
back. The plan is to release the
software for others to use as it
is further developed. ❚A HUGE 3D map depicts 11 billion
years of cosmic history and places
the tightest constraints ever on
our best model of the universe.
Captured by the Sloan Digital Sky
Survey (SDSS), it has bolstered our
leading picture of the cosmos,
even though it deepens one
enduring mystery.
Light travels at a finite speed,
so looking into space also means
peering back through time. This
new survey looks deep enough to
map 80 per cent of the universe’s
14-billion-year history. “There isn’t
anything else with that range of
coverage and that allows us to fill
this 11-billion-year gap between
the ancient and recent universe,”
says Kyle Dawson at the University
of Utah, who leads the extended
Baryon Oscillation Spectroscopic
Survey (eBOSS) team at SDSS.
The team observed galaxies
and quasars, which are the bright
centres of some galaxies, and used
their red shifts – changes in light
due to them moving away from
us – to measure distances and the
rate of expansion of the universe.
This lets us watch giant structures
such as galaxy clusters forming.
“The universe now is very
clumpy: there can be largethings like galaxies or planets in
one place, or nothing in another
place,” says Scott Dodelson at
Carnegie Mellon University in
Pennsylvania. That wasn’t always
the case. “It used to be that if you
went to one random place and
counted 1000 atoms and then
went to another random place,
you might count 1001 but
probably not even 1002.”Our leading approach to
understand how the universe
went from mostly homogeneous
to clumpy is a model called
lambda-CDM. Some past
measurements have hinted
that what we see in the universe
may not match that model’s
predictions, but the eBOSS map
shows no conflict at all (arxiv.org/
abs/2007.08991). So lambda-CDM
is holding up well.
The development of large-scale
structure is partly dependent on
the behaviour of particles knownas neutrinos in the early universe;
eBOSS was able to constrain their
mass, which is a big outstanding
problem in physics. It didn’t quite
nail it down, but the measurement
was as precise as the best ground-
based neutrino experiments.
The team also constrained the
shape of the universe 10 times
more tightly than our next best
set of observations. As predicted
by lambda-CDM, space-time as a
whole seems to be flat, not curved.
However, one existing conflict
has been exacerbated by the
survey. “Things are fitting
together remarkably well, with
the exception of the Hubble
constant,” says Wendy Freedman
at the University of Chicago.
This is a measure of the rate of
expansion of the universe. Our
two main ways of calculating
it – using the ancient cosmic
microwave background (CMB)
versus a local measurement of
the movement of nearby objects –
always disagree.
The eBOSS study agrees with the
CMB method, which deepens the
puzzle. “There’s probably some
missing physics somewhere, but
nobody has been able to come up
with it yet,” says Freedman. ❚“The system could be
adapted to help people
recognise individual birds
in their surroundings”Machine learning
The Sloan Digital Sky
Survey uses a 2.5-metre
telescope in New MexicoLeah CraneSLOAN
DI
GITAL
SKY^ SURVE
YNews
Best map of the universe created
It charts 11 billion years of the cosmos and deepens a long-standing mystery
AI learns to
recognise individual
birds from behindMichael Le Page