Mathematics,
Год журнала:
2024,
Номер
12(24), С. 3948 - 3948
Опубликована: Дек. 15, 2024
Cryptography
is
one
of
the
most
important
branches
information
security.
ensures
secure
communication
and
data
privacy,
it
has
been
increasingly
applied
in
healthcare
related
areas.
As
a
significant
cryptographic
method,
Hill
cipher
attracted
attention
from
experts
scholars.
To
enhance
security
traditional
(THC)
expand
its
application
medical
image
encryption,
novel
dynamic
with
Arnold
scrambling
technique
(DHCAST)
proposed
this
work.
Unlike
THC,
DHCAST
uses
time-varying
matrix
as
secret
key,
which
greatly
increases
new
successfully
images
encryption.
In
addition,
method
employs
Zeroing
Neural
Network
(ZNN)
decryption
to
find
inversion
key
(TVIKM).
order
efficiency
ZNN
for
solving
TVIKM,
fuzzy
zeroing
neural
network
(NFZNN)
model
constructed,
convergence
robustness
NFZNN
are
validated
by
both
theoretical
analysis
experiment
results.
Simulation
experiments
show
that
time
about
0.05
s,
while
(TZNN)
2
means
speed
400
times
TZNN
model.
Moreover,
Peak
Signal
Noise
Ratio
(PSNR)
Number
Pixel
Change
Rate
(NPCR)
algorithm
reach
9.51
99.74%,
respectively,
effectively
validates
excellent
encryption
quality
attack
prevention
ability.
Mathematics,
Год журнала:
2025,
Номер
13(2), С. 201 - 201
Опубликована: Янв. 9, 2025
In
this
paper,
we
first
design
the
corresponding
integration
algorithm
and
matlab
program
according
to
Gauss–Legendre
principle.
Then,
select
Lorenz
system,
Duffing
hidden
attractor
chaotic
system
Multi-wing
for
dynamics
analysis.
We
apply
integral
Runge–Kutta
solution
of
dissipative
systems
time
analyze
compare
differences
between
two
algorithms.
propose
a
basin
attraction
estimation
method
based
on
Lyapunov
exponent
decision
criterion
method.
This
can
better
obtain
region
distinguish
pseudo-attractor,
which
provides
new
way
Finally,
use
FPGA
technology
realize
four
algorithm.
Electronics,
Год журнала:
2025,
Номер
14(4), С. 766 - 766
Опубликована: Фев. 16, 2025
Locally
active
memristors
with
an
Edge-of-Chaos
kernel
(EOCK)
represent
a
significant
advancement
in
the
simulation
of
neuromorphic
dynamics.
However,
current
research
on
EOCK
remains
at
circuit
level,
without
further
analysis
their
feasibility.
In
this
context,
we
designed
memristor
and
installed
it
third-order
circuit,
where
showed
local
activity
stability
under
defined
voltage
inductance
parameters.
This
behavior
ensured
that
by
varying
input
inductance,
could
effectively
simulate
various
neural
activities,
including
inhibitory
postsynaptic
potential
chaotic
waveforms.
By
subsequently
integrating
into
Hopfield
network
(HNN)
framework
substituting
self-coupling
weight,
observed
rich
spectrum
dynamic
behaviors,
rare
phenomenon
antimonotonicity
bubble
bifurcation.
Finally,
used
hardware
circuits
to
realize
these
generated
phenomena,
confirming
feasibility
memristor.
introducing
HNN
studying
its
implementation,
study
provides
theoretical
insights
empirical
basis
for
developing
systems
replicate
complexity
human
brain
functions.
reference
development
application
future.
Mathematics,
Год журнала:
2025,
Номер
13(5), С. 726 - 726
Опубликована: Фев. 24, 2025
In
comparison
with
dissipative
chaos,
conservative
chaos
is
better
equipped
to
handle
the
risks
associated
reconstruction
of
phase
space
due
absence
attractors.
This
paper
proposes
a
novel
five-dimensional
(5D)
memristive
hyperchaotic
system
(CMHS),
by
incorporating
memristors
into
four-dimensional
(4D)
chaotic
(CCS).
We
conducted
comprehensive
analysis,
using
Lyapunov
exponent
diagrams,
bifurcation
portraits,
equilibrium
points,
and
spectral
entropy
maps
thoroughly
verify
system’s
properties.
The
exhibited
characteristics
such
as
hyperchaos
multi-stability
over
an
ultra-wide
range
parameters
initial
values,
accompanied
transient
quasi-periodic
phenomena.
Subsequently,
pseudorandom
sequences
generated
new
were
tested
demonstrated
excellent
performance,
passing
all
tests
set
National
Institute
Standards
Technology
(NIST).
final
stage
research,
image-encryption
application
based
on
5D
CMHS
was
proposed.
Through
security
feasibility
algorithm
confirmed.
Mathematics,
Год журнала:
2025,
Номер
13(5), С. 787 - 787
Опубликована: Фев. 27, 2025
In
deep
learning,
convolutional
layers
typically
bear
the
majority
of
computational
workload
and
are
often
primary
contributors
to
performance
bottlenecks.
The
widely
used
convolution
algorithm
is
based
on
IM2COL
transform
take
advantage
highly
optimized
GEMM
(General
Matrix
Multiplication)
kernel
acceleration,
using
BLAS
(Basic
Linear
Algebra
Subroutine)
library,
which
tends
incur
additional
memory
overhead.
Recent
studies
have
indicated
that
direct
approaches
can
outperform
traditional
implementations
without
this
paper,
we
propose
a
high-performance
implementation
for
inference
preserves
channel-first
data
layout
layer
inputs/outputs.
We
evaluate
our
proposed
multi-core
ARM
CPU
platform
compare
it
with
state-of-the-art
optimization
techniques.
Experimental
results
demonstrate
new
performs
better
across
evaluated
scenarios
platforms.
