快猫短视频

BECAUSE computers work in a very different way from the brain, it鈥檚 hard to
get them to do many of the things that the brain does effortlessly, such as
recognising handwriting or spotting a face in a crowd. But a team of Swiss and
American researchers has found a way around this problem by building an
electronic circuit that mimics the way the brain works. They believe similar
circuits could outpace a supercomputer on specialised tasks.

Designing circuits that work on the same principle as the brain is a
challenge鈥攏ot least because neuroscientists are not exactly sure how the
brain processes information. 鈥淭he brain can do incredible things that computers
aren鈥檛 able to, but how the brain does it isn鈥檛 obvious,鈥� says Richard
Hahnloser, who built the circuit at the Institute of Neuroinformatics in
Zurich.

But scientists do have some leads to go on. One key factor is that the brain
can simultaneously process digital and analogue signals. 快猫短视频s believe that
special links between neurons, known as recurrent connections, help the brain to
do this. When a neuron fires, recurrent connections pump signals into
neighbouring neurons. This causes the neighbours to fire, though less strongly,
sending reduced signals to all their neighbours, including the original neuron.
In this way the original signal is spread out and amplified at the same
time.

To emulate this, the researchers built a circuit in the shape of a ring of
electronic neurons. Each neuron is linked to its four nearest neighbours with a
recurrent connection. So when a signal is fed into one neuron, the signal
spreads via neighbouring neurons around the ring and amplifies the signal from
the first neuron.

To control this process, the team added a central 鈥渋nhibitory neuron鈥�. This
adds up all the signals across the whole circuit and uses the result to send an
appropriate signal back to all the neurons to reduce their output. Neurons
firing only weakly are effectively turned off, while others are weakened. As a
result, amplification is restricted to a more local group of neurons.

When the researchers tested the circuit, they found it processed information
much like the brain. In particular, when multiple signals were fed in, the
inhibitory neuron suppressed the weaker signals. This is akin to 鈥渟ensory
attention鈥� that occurs, say, when you focus on an object, and everything else
fades into the background.

According to Hahnloser, now at the Massachusetts Institute of Technology,
circuits like this will out-perform standard circuits in dedicated tasks such as
object recognition. 鈥淲hat is so different here is that they use much less power
and they can be much faster because they can be designed for specific tasks,鈥� he
says. Andrew Derrington, who studies brain function at the University of
Nottingham, is impressed. 鈥淭o do this in silicon and show that it works is
really what we鈥檝e needed.鈥�