Abstract
In
the
central
nervous
system,
sequences
of
neural
activity
form
trajectories
on
low
dimensional
manifolds.
The
computation
underlying
flexible
cognition
and
behavior
relies
dynamic
control
these
structures.
For
example
different
tasks
or
behaviors
are
represented
subspaces,
requiring
fast
timescale
subspace
rotation
to
move
from
one
next.
flexibility
in
a
particular
behavior,
trajectory
must
be
dynamically
controllable
within
that
behaviorally
determined
subspace.
To
understand
how
their
subspaces
may
implemented
circuits,
we
first
characterized
relationship
between
features
aspects
projection.
Based
this,
propose
mechanisms
can
act
local
circuits
modulate
thereby
controlling
subspaces.
particular,
show
gain
modulation
transient
synaptic
currents
speed
path
clustered
inhibition
determines
manifold
orientation.
Together,
enable
substrate
for
bioRxiv (Cold Spring Harbor Laboratory),
Journal Year:
2023,
Volume and Issue:
unknown
Published: Sept. 23, 2023
Sensory
stimuli
associated
with
aversive
outcomes
cause
multiple
behavioral
responses
related
to
an
animal's
evolving
emotional
state,
but
neural
mechanisms
underlying
these
processes
remain
unclear.
Here
were
presented
mice,
eliciting
two
reflecting
fear
and
flight
safety:
tremble
ingress
into
a
virtual
burrow.
Inactivation
of
basolateral
amygdala
(BLA)
eliminated
differential
neutral
without
eliminating
themselves,
suggesting
BLA
signals
valence,
not
motor
commands.
However,
two-photon
imaging
revealed
that
neurons
typically
exhibited
mixed
selectivity
for
stimulus
identity,
and/or
ingress.
Despite
heterogeneous
selectivity,
representational
geometry
was
lower-dimensional
when
encoding
safety,
enabling
generalization
emotions
across
conditions.
Further,
valence
coding
directions
orthogonal,
allowing
linear
readouts
specialize.
Thus
confers
computational
properties
identify
specialized
circuits
variables
describing
states:
conditions,
lacking
interference
from
other
readouts.
Object
classification
has
been
proposed
as
a
principal
objective
of
the
primate
ventral
visual
stream
and
used
an
optimization
target
for
deep
neural
network
models
(DNNs)
system.
However,
brain
areas
represent
many
different
types
information,
optimizing
object
identity
alone
does
not
constrain
how
other
information
may
be
encoded
in
representations.
Information
about
scene
parameters
discarded
altogether
('invariance'),
represented
non-interfering
subspaces
population
activity
('factorization')
or
entangled
fashion.
In
this
work,
we
provide
evidence
that
factorization
is
normative
principle
biological
monkey
hierarchy,
found
pose
background
from
increased
higher-level
regions
strongly
contributed
to
improving
decoding
performance.
We
then
conducted
large-scale
analysis
individual
-
lighting,
background,
camera
viewpoint,
diverse
library
DNN
Models
which
best
matched
neural,
fMRI,
behavioral
data
both
monkeys
humans
across
12
datasets
tended
those
factorized
most
strongly.
Notably,
invariance
these
was
consistently
associated
with
matches
data,
suggesting
maintaining
non-class
often
preferred
dropping
it
altogether.
Thus,
propose
widely
strategy
brains
thereof.
bioRxiv (Cold Spring Harbor Laboratory),
Journal Year:
2024,
Volume and Issue:
unknown
Published: May 4, 2024
Abstract
The
dorsal
CA2
subregion
(dCA2)
of
the
hippocampus
exerts
a
critical
role
in
social
novelty
recognition
(SNR)
memory
and
promotion
aggression.
Whether
aggression
SNR
functions
dCA2
are
related
or
represent
independent
processes
is
unknown.
Here
we
investigated
hypotheses
that
an
animal
more
likely
to
attack
novel
compared
familiar
promotes
through
its
ability
discriminate
between
conspecifics.
To
test
these
ideas,
conducted
multi-day
resident
intruder
(R-I)
towards
We
found
mice
were
familiarized
silencing
caused
profound
inhibition
than
intruder.
explore
whether
how
pyramidal
neurons
encode
aggression,
recorded
their
activity
using
microendoscopic
calcium
imaging
throughout
days
R-I
test.
fraction
selectively
activated
inhibited
during
exploration,
dominance,
behaviors
signals
enhanced
interaction
with
conspecific.
Based
on
population
activity,
set
binary
linear
classifiers
accurately
decoded
was
engaged
each
forms
behavior.
Of
particular
interest,
accuracy
decoding
greater
intruders,
significant
cross-day
same
day
but
not
for
familiar-novel
pair.
Together,
findings
demonstrate
integrates
information
about
behavioral
state
promote
Object
classification
has
been
proposed
as
a
principal
objective
of
the
primate
ventral
visual
stream
and
used
an
optimization
target
for
deep
neural
network
models
(DNNs)
system.
However,
brain
areas
represent
many
different
types
information,
optimizing
object
identity
alone
does
not
constrain
how
other
information
may
be
encoded
in
representations.
Information
about
scene
parameters
discarded
altogether
(‘invariance’),
represented
non-interfering
subspaces
population
activity
(‘factorization’)
or
entangled
fashion.
In
this
work,
we
provide
evidence
that
factorization
is
normative
principle
biological
monkey
hierarchy,
found
pose
background
from
increased
higher-level
regions
strongly
contributed
to
improving
decoding
performance.
We
then
conducted
large-scale
analysis
individual
–
lighting,
background,
camera
viewpoint,
diverse
library
DNN
Models
which
best
matched
neural,
fMRI,
behavioral
data
both
monkeys
humans
across
12
datasets
tended
those
factorized
most
strongly.
Notably,
invariance
these
was
consistently
associated
with
matches
data,
suggesting
maintaining
non-class
often
preferred
dropping
it
altogether.
Thus,
propose
widely
strategy
brains
thereof.
Abstract
In
the
central
nervous
system,
sequences
of
neural
activity
form
trajectories
on
low
dimensional
manifolds.
The
computation
underlying
flexible
cognition
and
behavior
relies
dynamic
control
these
structures.
For
example
different
tasks
or
behaviors
are
represented
subspaces,
requiring
fast
timescale
subspace
rotation
to
move
from
one
next.
flexibility
in
a
particular
behavior,
trajectory
must
be
dynamically
controllable
within
that
behaviorally
determined
subspace.
To
understand
how
their
subspaces
may
implemented
circuits,
we
first
characterized
relationship
between
features
aspects
projection.
Based
this,
propose
mechanisms
can
act
local
circuits
modulate
thereby
controlling
subspaces.
particular,
show
gain
modulation
transient
synaptic
currents
speed
path
clustered
inhibition
determines
manifold
orientation.
Together,
enable
substrate
for