Nature Communications,
Journal Year:
2022,
Volume and Issue:
13(1)
Published: Jan. 10, 2022
There
are
two
principle
approaches
for
learning
in
artificial
intelligence:
error-driven
global
and
neuroscience-oriented
local
learning.
Integrating
them
into
one
network
may
provide
complementary
capabilities
versatile
scenarios.
At
the
same
time,
neuromorphic
computing
holds
great
promise,
but
still
needs
plenty
of
useful
algorithms
algorithm-hardware
co-designs
to
fully
exploit
its
advantages.
Here,
we
present
a
global-local
synergic
model
by
introducing
brain-inspired
meta-learning
paradigm
differentiable
spiking
incorporating
neuronal
dynamics
synaptic
plasticity.
It
can
meta-learn
plasticity
receive
top-down
supervision
information
multiscale
We
demonstrate
advantages
this
multiple
different
tasks,
including
few-shot
learning,
continual
fault-tolerance
vision
sensors.
achieves
significantly
higher
performance
than
single-learning
methods.
further
implement
Tianjic
platform
exploiting
prove
that
utilize
many-core
architecture
develop
hybrid
computation
paradigm.
Philosophical Transactions of the Royal Society B Biological Sciences,
Journal Year:
2013,
Volume and Issue:
369(1633), P. 20130288 - 20130288
Published: Dec. 3, 2013
The
synaptic
plasticity
and
memory
hypothesis
asserts
that
activity-dependent
is
induced
at
appropriate
synapses
during
formation
both
necessary
sufficient
for
the
encoding
trace
storage
of
type
mediated
by
brain
area
in
which
it
observed.
Criteria
establishing
necessity
sufficiency
such
mediating
have
been
identified
are
here
reviewed
relation
to
new
work
using
some
diverse
techniques
contemporary
neuroscience.
Evidence
derived
optical
imaging,
molecular-genetic
optogenetic
conjunction
with
behavioural
analyses
continues
offer
support
idea
changing
strength
connections
between
neurons
one
major
mechanisms
engrams
stored
brain.
The MIT Press eBooks,
Journal Year:
2015,
Volume and Issue:
unknown
Published: May 22, 2015
Two
distinguished
neuroscientists
distil
general
principles
from
more
than
a
century
of
scientific
study,
“reverse
engineering”
the
brain
to
understand
its
design.
Neuroscience
research
has
exploded,
with
fifty
thousand
applying
increasingly
advanced
methods.
A
mountain
new
facts
and
mechanisms
emerged.
And
yet
principled
framework
organize
this
knowledge
been
missing.
In
book,
Peter
Sterling
Simon
Laughlin,
two
leading
neuroscientists,
strive
fill
gap,
outlining
set
organizing
explain
whys
neural
design
that
allow
compute
so
efficiently.
Setting
out
engineer”
brain—disassembling
it
it—Sterling
Laughlin
first
consider
why
an
animal
should
need
brain,
tracing
computational
abilities
bacterium
protozoan
worm.
They
examine
bigger
brains
advantages
“anticipatory
regulation”;
identify
constraints
on
“nanofy”;
demonstrate
routes
efficiency
in
integrated
molecular
system,
phototransduction.
show
at
finer
scales
lower
levels
apply
larger
higher
levels;
describe
wiring
efficiency;
discuss
learning
as
principle
biological
includes
“save
only
what
is
needed.”
avoid
speculation
about
how
might
work
endeavor
make
sense
already
known.
Their
distinctive
contribution
gather
coherent
basic
rules
exemplify
them
across
spatial
functional
scales.
Frontiers in Neural Circuits,
Journal Year:
2016,
Volume and Issue:
9
Published: Jan. 19, 2016
Classical
Hebbian
learning
puts
the
emphasis
on
joint
pre-
and
postsynaptic
activity,
but
neglects
potential
role
of
neuromodulators.
Since
neuromodulators
convey
information
about
novelty
or
reward,
influence
synaptic
plasticity
is
useful
not
just
for
action
in
classical
conditioning,
also
to
decide
"when"
create
new
memories
response
a
flow
sensory
stimuli.
In
this
review,
we
focus
timing
requirements
activity
conjunction
with
one
several
phasic
neuromodulatory
signals.
While
text
conceptual
models
mathematical
theories,
discuss
some
experimental
evidence
neuromodulation
Spike-Timing-Dependent
Plasticity.
We
highlight
importance
mechanisms
bridging
temporal
gap
between
stimulation
signals,
develop
framework
class
neo-Hebbian
three-factor
rules
that
depend
presynaptic
variables
as
well
