While
many
brain
networks
are
specialised
for
processing
specific
types
of
information,
a
network
frontoparietal
regions
is
engaged
by
wide
range
cognitive
demands.
Here
we
review
recent
work
highlighting
the
flexibility
information
coding
in
these
regions,
including
their
potential
to
differentiate
variety
different
and
dynamic
selectivity
that
currently
relevant.
But
does
all
decodable
activity
constitute
behaviourally
meaningful
brain?
Examining
emerging
methods,
find
direct
link
behaviour
can
be
made
some,
but
not
all,
information.
The
data
suggest
flexible
resource
suitable
creating
temporary,
arbitrary,
associations
between
aspects
needed
each
task.
However,
tighter
field-wide
focus
on
decoding-behaviour
relationships
specify
how
this
gives
rise
astounding
human
capacity
thought
action.
Brain Behavior and Immunity,
Journal Year:
2023,
Volume and Issue:
115, P. 470 - 479
Published: Nov. 14, 2023
Artificial
intelligence
(AI)
is
often
used
to
describe
the
automation
of
complex
tasks
that
we
would
attribute
to.
Machine
learning
(ML)
commonly
understood
as
a
set
methods
develop
an
AI.
Both
have
seen
recent
boom
in
usage,
both
scientific
and
commercial
fields.
For
community,
ML
can
solve
bottle
necks
created
by
complex,
multi-dimensional
data
generated,
for
example,
functional
brain
imaging
or
*omics
approaches.
here
identify
patterns
could
not
been
found
using
traditional
statistic
However,
comes
with
serious
limitations
need
be
kept
mind:
their
tendency
optimise
solutions
input
means
it
crucial
importance
externally
validate
any
findings
before
considering
them
more
than
hypothesis.
Their
black-box
nature
implies
decisions
usually
cannot
understood,
which
renders
use
medical
decision
making
problematic
lead
ethical
issues.
Here,
present
introduction
curious
field
ML/AI.
We
explain
principles
well
methodological
advancements
discuss
risks
what
see
future
directions
field.
Finally,
show
practical
examples
neuroscience
illustrate
ML.
Nature Neuroscience,
Journal Year:
2024,
Volume and Issue:
27(5), P. 988 - 999
Published: March 18, 2024
Abstract
A
fundamental
human
cognitive
feat
is
to
interpret
linguistic
instructions
in
order
perform
novel
tasks
without
explicit
task
experience.
Yet,
the
neural
computations
that
might
be
used
accomplish
this
remain
poorly
understood.
We
use
advances
natural
language
processing
create
a
model
of
generalization
based
on
instructions.
Models
are
trained
set
common
psychophysical
tasks,
and
receive
embedded
by
pretrained
model.
Our
best
models
can
previously
unseen
with
an
average
performance
83%
correct
solely
(that
is,
zero-shot
learning).
found
scaffolds
sensorimotor
representations
such
activity
for
interrelated
shares
geometry
semantic
instructions,
allowing
cue
proper
composition
practiced
skills
settings.
show
how
generates
description
it
has
identified
using
only
motor
feedback,
which
subsequently
guide
partner
task.
offer
several
experimentally
testable
predictions
outlining
information
must
represented
facilitate
flexible
general
cognition
brain.
Thalamocortical
interaction
is
a
ubiquitous
functional
motif
in
the
mammalian
brain.
Previously
(Hwang
et
al.,
2021),
we
reported
that
lesions
to
network
hubs
human
thalamus
are
associated
with
multi-domain
behavioral
impairments
language,
memory,
and
executive
functions.
Here,
show
how
task-evoked
thalamic
activity
organized
support
these
broad
cognitive
abilities.
We
analyzed
magnetic
resonance
imaging
(MRI)
data
from
subjects
performed
127
tasks
encompassing
range
of
representations.
first
investigated
spatial
organization
found
basis
set
patterns
evoked
processing
needs
each
task.
Specifically,
anterior,
medial,
posterior-medial
exhibit
hub-like
profiles
suggestive
participation.
These
task
overlapped
interlinking
cortical
systems.
To
further
determine
relevance
thalamocortical
connectivity,
built
data-driven
model
test
whether
can
be
used
predict
activity.
The
predicted
task-specific
patterns,
outperformed
comparison
models
on
cortical,
hippocampal,
striatal
regions.
Simulated
low-dimensional,
multi-task
hub
regions
impaired
prediction.
This
simulation
result
was
supported
by
neuropsychological
patients
focal
lesions.
In
summary,
our
results
suggest
general
organizational
principle
system
supports
The Neuroscientist,
Journal Year:
2022,
Volume and Issue:
30(3), P. 367 - 377
Published: Oct. 17, 2022
The
human
brain
is
composed
of
multiple,
discrete,
functionally
specialized
regions
that
are
interconnected
to
form
large-scale
distributed
networks.
Using
advanced
brain-imaging
methods
and
machine-learning
analytical
approaches,
recent
studies
have
demonstrated
regional
activity
during
the
performance
various
cognitive
tasks
can
be
accurately
predicted
from
patterns
task-independent
connectivity.
In
this
review
article,
we
first
present
evidence
for
predictability
structural
connectivity
(i.e.,
white
matter
connections)
functional
temporally
synchronized
task-free
activations).
We
then
discuss
implications
such
predictions
clinical
populations,
as
patients
diagnosed
with
psychiatric
disorders
or
neurologic
diseases,
study
brain-behavior
associations.
conclude
may
serve
an
infrastructure
dictates
activity,
pinpoint
several
open
questions
directions
future
research.
bioRxiv (Cold Spring Harbor Laboratory),
Journal Year:
2023,
Volume and Issue:
unknown
Published: June 29, 2023
During
cognitive
task
learning,
neural
representations
must
be
rapidly
constructed
for
novel
performance,
then
optimized
robust
practiced
performance.
How
the
geometry
of
changes
to
enable
this
transition
from
performance
remains
unknown.
We
hypothesized
that
practice
involves
a
shift
compositional
(task-general
activity
patterns
can
flexibly
reused
across
tasks)
conjunctive
(task-specific
specialized
current
task).
Functional
MRI
during
learning
multiple
complex
tasks
substantiated
dynamic
representations,
which
was
associated
with
reduced
cross-task
interference
(via
pattern
separation)
and
behavioral
improvement.
Further,
we
found
conjunctions
originated
in
subcortex
(hippocampus
cerebellum)
slowly
spread
cortex,
extending
memory
systems
theories
encompass
representation
learning.
The
formation
hence
serves
as
computational
signature
reflecting
cortical-subcortical
dynamics
optimize
human
brain.
Nature Communications,
Journal Year:
2024,
Volume and Issue:
15(1)
Published: Aug. 1, 2024
Behavioral
flexibility
relies
on
the
brain's
ability
to
switch
rapidly
between
multiple
tasks,
even
when
task
rule
is
not
explicitly
cued
but
must
be
inferred
through
trial
and
error.
The
underlying
neural
circuit
mechanism
remains
poorly
understood.
We
investigated
recurrent
networks
(RNNs)
trained
perform
an
analog
of
classic
Wisconsin
Card
Sorting
Test.
consist
two
modules
responsible
for
representation
sensorimotor
mapping,
respectively,
where
each
module
comprised
a
with
excitatory
neurons
three
major
types
inhibitory
neurons.
found
that
by
self-sustained
persistent
activity
across
trials,
error
monitoring
gated
mapping
emerged
from
training.
Systematic
dissection
RNNs
revealed
detailed
consistent
different
hyperparameters.
networks'
dynamical
trajectories
rules
resided
in
separate
subspaces
population
activity;
collapsed
performance
was
reduced
chance
level
dendrite-targeting
somatostatin-expressing
interneurons
were
silenced,
illustrating
how
phenomenological
description
representational
explained
specific
mechanism.
flexible
switching
unclear.
Here
authors
analyzed
modular
network
models
cell
reveal
uncued
switching.
