Proceedings of the National Academy of Sciences,
Год журнала:
2025,
Номер
122(10)
Опубликована: Март 5, 2025
Despite
the
impressive
performance
of
biological
and
artificial
networks,
an
intuitive
understanding
how
their
local
learning
dynamics
contribute
to
network-level
task
solutions
remains
a
challenge
this
date.
Efforts
bring
more
scale
indeed
lead
valuable
insights,
however,
general
constructive
approach
describe
goals
that
is
both
interpretable
adaptable
across
diverse
tasks
still
missing.
We
have
previously
formulated
information
processing
goal
highly
for
model
neuron
with
compartmental
structure.
Building
on
recent
advances
in
Partial
Information
Decomposition
(PID),
we
here
derive
corresponding
parametric
rule,
which
allows
us
introduce
"infomorphic"
neural
networks.
demonstrate
versatility
these
networks
perform
from
supervised,
unsupervised,
memory
learning.
By
leveraging
nature
PID
framework,
infomorphic
represent
tool
advance
our
intricate
structure
Trends in Neurosciences,
Год журнала:
2023,
Номер
46(5), С. 377 - 393
Опубликована: Март 27, 2023
Crossmodal
plasticity
is
a
textbook
example
of
the
ability
brain
to
reorganize
based
on
use.
We
review
evidence
from
auditory
system
showing
that
such
reorganization
has
significant
limits,
dependent
pre-existing
circuitry
and
top-down
interactions,
extensive
often
absent.
argue
does
not
support
hypothesis
crossmodal
responsible
for
closing
critical
periods
in
deafness,
instead
represents
neuronal
process
dynamically
adaptable.
evaluate
changes
both
developmental
adult-onset
which
start
as
early
mild-moderate
hearing
loss
show
reversibility
when
restored.
Finally,
appear
affect
preconditions
successful
restoration.
Given
its
dynamic
versatile
nature,
we
describe
how
this
can
be
exploited
improving
clinical
outcomes
after
neurosensory
Neuron,
Год журнала:
2023,
Номер
111(18), С. 2918 - 2928.e8
Опубликована: Сен. 1, 2023
Predictive
processing
postulates
the
existence
of
prediction
error
neurons
in
cortex.
Neurons
with
both
negative
and
positive
response
properties
have
been
identified
layer
2/3
visual
cortex,
but
whether
they
correspond
to
transcriptionally
defined
subpopulations
is
unclear.
Here
we
used
activity-dependent,
photoconvertible
marker
CaMPARI2
tag
mouse
cortex
during
stimuli
behaviors
designed
evoke
errors.
We
performed
single-cell
RNA-sequencing
on
these
populations
found
that
previously
annotated
Adamts2
Rrad
transcriptional
cell
types
were
enriched
when
photolabeling
drive
or
responses,
respectively.
Finally,
validated
results
functionally
by
designing
artificial
promoters
for
use
AAV
vectors
express
genetically
encoded
calcium
indicators.
Thus,
distinct
can
be
targeted
using
exhibit
distinguishable
responses.
Trends in Neurosciences,
Год журнала:
2023,
Номер
47(2), С. 92 - 105
Опубликована: Дек. 14, 2023
Predictive
processing
models
posit
that
brains
constantly
attempt
to
predict
their
sensory
inputs.
Prediction
errors
signal
when
these
predictions
are
incorrect
and
thought
be
instructive
signals
drive
corrective
plasticity.
Recent
findings
support
the
idea
locus
coeruleus
(LC)
-
a
brain-wide
neuromodulatory
system
several
types
of
prediction
error.
I
discuss
how
proposing
LC
global
model
failures:
instances
where
about
world
strongly
violated.
Focusing
on
cortex,
explore
utility
this
in
learning
rate
control,
circuit
may
compute
signal,
view
aid
our
understanding
neurodivergence.
How
is
the
massive
dimensionality
and
complexity
of
microscopic
constituents
nervous
system
brought
under
sufficiently
tight
control
so
as
to
coordinate
adaptive
behaviour?
A
powerful
means
for
striking
this
balance
poise
neurons
close
critical
point
a
phase
transition,
at
which
small
change
in
neuronal
excitability
can
manifest
nonlinear
augmentation
activity.
brain
could
mediate
transition
key
open
question
neuroscience.
Here,
I
propose
that
different
arms
ascending
arousal
provide
with
diverse
set
heterogeneous
parameters
be
used
modulate
receptivity
target
neurons-in
other
words,
act
mediating
order.
Through
series
worked
examples,
demonstrate
how
neuromodulatory
interact
inherent
topological
subsystems
complex
behaviour.
Current Opinion in Neurobiology,
Год журнала:
2024,
Номер
85, С. 102853 - 102853
Опубликована: Фев. 22, 2024
The
brain
is
a
remarkably
capable
and
efficient
system.
It
can
process
store
huge
amounts
of
noisy
unstructured
information
using
minimal
energy.
In
contrast,
current
artificial
intelligence
(AI)
systems
require
vast
resources
for
training
while
still
struggling
to
compete
in
tasks
that
are
trivial
biological
agents.
Thus,
brain-inspired
engineering
has
emerged
as
promising
new
avenue
designing
sustainable,
next-generation
AI
systems.
Here,
we
describe
how
dendritic
mechanisms
neurons
have
inspired
innovative
solutions
significant
problems,
including
credit
assignment
multilayer
networks,
catastrophic
forgetting,
high
energy
consumption.
These
findings
provide
exciting
alternatives
existing
architectures,
showing
research
pave
the
way
building
more
powerful
energy-efficient
learning
PLoS Computational Biology,
Год журнала:
2024,
Номер
20(4), С. e1011183 - e1011183
Опубликована: Апрель 1, 2024
One
of
the
key
problems
brain
faces
is
inferring
state
world
from
a
sequence
dynamically
changing
stimuli,
and
it
not
yet
clear
how
sensory
system
achieves
this
task.
A
well-established
computational
framework
for
describing
perceptual
processes
in
provided
by
theory
predictive
coding.
Although
original
proposals
coding
have
discussed
temporal
prediction,
later
work
developing
mostly
focused
on
static
questions
neural
implementation
properties
networks
remain
open.
Here,
we
address
these
present
formulation
model
that
can
be
naturally
implemented
recurrent
networks,
which
activity
dynamics
rely
only
local
inputs
to
neurons,
learning
utilises
Hebbian
plasticity.
Additionally,
show
approximate
performance
Kalman
filter
predicting
behaviour
linear
systems,
behave
as
variant
does
track
its
own
subjective
posterior
variance.
Importantly,
achieve
similar
accuracy
without
performing
complex
mathematical
operations,
but
just
employing
simple
computations
biological
networks.
Moreover,
when
trained
with
natural
dynamic
inputs,
found
produce
Gabor-like,
motion-sensitive
receptive
fields
resembling
those
observed
real
neurons
visual
areas.
In
addition,
demonstrate
effectively
generalized
nonlinear
systems.
Overall,
models
presented
paper
biologically
plausible
circuits
predict
future
stimuli
may
guide
research
understanding
specific
areas
involved
prediction.
Non-invasive
recordings
of
gross
neural
activity
in
humans
often
show
responses
to
omitted
stimuli
steady
trains
identical
stimuli.
This
has
been
taken
as
evidence
for
the
coding
prediction
or
error.
However,
such
omission
from
invasive
cellular-scale
animal
models
is
scarce.
Here,
we
sought
characterise
using
extracellular
auditory
cortex
anaesthetised
rats.
We
profiled
across
local
field
potentials
(LFP),
analogue
multiunit
(AMUA),
and
single/multi-unit
spiking
activity,
that
were
fixed-rate
acoustic
noise
bursts
where
5%
randomly
omitted.
Current Research in Neurobiology,
Год журнала:
2024,
Номер
6, С. 100129 - 100129
Опубликована: Янв. 1, 2024
We
argue
that
prediction
success
maximization
is
a
basic
objective
of
cognition
and
cortex,
it
compatible
with
but
distinct
from
error
minimization,
neither
requires
subtractive
coding,
there
clear
neurobiological
evidence
for
the
amplification
predicted
signals,
we
are
unconvinced
by
proposed
in
support
coding.
outline
recent
discoveries
showing
pyramidal
cells
on
which
our
cognitive
capabilities
depend
usually
transmit
information
about
input
to
their
basal
dendrites
amplify
transmission
when
distal
apical
provides
context
agrees
feedforward
both
depolarizing,
i.e.,
excitatory
rather
than
inhibitory.
Though
these
intracellular
require
level
technical
expertise
beyond
current
abilities
most
neuroscience
labs,
they
not
controversial
acclaimed
as
groundbreaking.
note
this
cellular
cooperative
context-sensitivity
greatly
enhances
mammalian
neocortex,
much
remains
be
discovered
concerning
its
evolution,
development,
pathology.
Non-linear
summation
of
synaptic
inputs
to
the
dendrites
pyramidal
neurons
has
been
proposed
increase
computation
capacity
through
coincidence
detection,
signal
amplification,
and
additional
logic
operations
such
as
XOR.
Supralinear
dendritic
integration
documented
extensively
in
principal
neurons,
mediated
by
several
voltage-dependent
conductances.
It
also
reported
parvalbumin-positive
hippocampal
basket
cells,
innervated
feedback
excitatory
synapses.
Whether
other
interneurons,
which
support
feed-forward
or
inhibition
neuron
dendrites,
exhibit
local
non-linear
excitation
is
not
known.
Here,
we
use
patch-clamp
electrophysiology,
two-photon
calcium
imaging
glutamate
uncaging,
show
that
supralinear
near-synchronous
spatially
clustered
glutamate-receptor
depolarization
occurs
NDNF-positive
neurogliaform
cells
oriens-lacunosum
moleculare
interneurons
mouse
hippocampus.
was
detected
via
recordings
somatic
depolarizations
elicited
uncaging
on
fragments,
and,
concurrent
transients.
Supralinearity
abolished
blocking
NMDA
receptors
(NMDARs)
but
resisted
blockade
voltage-gated
sodium
channels.
Blocking
L-type
channels
signalling
only
had
a
minor
effect
voltage
supralinearity.
Dendritic
boosting
signals
argues
for
previously
unappreciated
computational
complexity
dendrite-projecting
inhibitory