Electronics,
Journal Year:
2024,
Volume and Issue:
13(24), P. 5042 - 5042
Published: Dec. 22, 2024
Federated
learning
(FL)
strikes
a
balance
between
privacy
preservation
and
collaborative
model
training.
However,
the
periodic
transmission
of
updates
or
parameters
from
each
client
to
federated
server
incurs
substantial
communication
overhead,
especially
for
participants
with
limited
network
bandwidth.
This
overhead
significantly
hampers
practical
applicability
FL
in
real-world
scenarios.
To
address
this
challenge,
we
propose
FedSparse,
an
innovative
sparse
framework
designed
enhance
efficiency.
The
core
idea
behind
FedSparse
is
introduce
regularization
term
into
client’s
objective
function,
thereby
reducing
number
that
need
be
transmitted.
incorporates
Resource
Optimization
Proximal
(ROP)
Importance-based
Regularization
Weighting
(IRW)
mechanism
update
function.
local
process
optimizes
both
empirical
risk
by
applying
weighted
importance.
By
making
minimal
modifications
traditional
approaches,
effectively
reduces
transmitted,
decreasing
overhead.
We
evaluate
effectiveness
through
experiments
on
various
datasets
under
non-independent
identically
distributed
(non-IID)
conditions,
demonstrating
its
flexibility
resource-constrained
environments.
On
MNIST,
Fashion-MNIST,
CIFAR
datasets,
24%,
17%,
5%,
respectively,
compared
baseline
algorithm,
while
maintaining
similar
performance.
Additionally,
simulated
non-IID
achieves
6%
8%
reduction
resource
consumption.
adjusting
sparsity
intensity
hyperparameter,
demonstrate
can
tailored
different
applications
varying
constraints.
Finally,
ablation
studies
highlight
individual
contributions
ROP
IRW
modules
overall
improvement
Scientific Reports,
Journal Year:
2025,
Volume and Issue:
15(1)
Published: Jan. 2, 2025
This
study
presents
a
novel
privacy-preserving
self-supervised
(SSL)
framework
for
COVID-19
classification
from
lung
CT
scans,
utilizing
federated
learning
(FL)
enhanced
with
Paillier
homomorphic
encryption
(PHE)
to
prevent
third-party
attacks
during
training.
The
FL-SSL
based
employs
two
publicly
available
scan
datasets
which
are
considered
as
labeled
and
an
unlabeled
dataset.
dataset
is
split
into
three
subsets
assumed
be
collected
hospitals.
Training
done
using
the
Bootstrap
Your
Own
Latent
(BYOL)
contrastive
SSL
VGG19
encoder
followed
by
attention
CNN
blocks
(VGG19
+
CNN).
input
processed
selecting
largest
portion
of
each
automated
selection
approach
64
×
size
utilized
reduce
computational
complexity.
Healthcare
privacy
issues
addressed
collaborative
training
across
decentralized
secure
aggregation
PHE,
underscoring
effectiveness
this
approach.
Three
used
train
local
BYOL
model,
together
optimizes
central
encoder.
employed
(updated
CNN),
resulting
in
accuracy
97.19%,
precision
97.43%,
recall
98.18%.
reliability
framework's
performance
demonstrated
through
statistical
analysis
five-fold
cross-validation.
efficacy
proposed
further
showcased
showing
its
on
distinct
modality
datasets:
skin
cancer,
breast
chest
X-rays.
In
conclusion,
offers
promising
solution
accurate
diagnosis
X-rays,
preserving
overcoming
challenges
scarcity
Future Internet,
Journal Year:
2025,
Volume and Issue:
17(1), P. 18 - 18
Published: Jan. 5, 2025
Federated
Learning
(FL)
is
a
distributed
machine-learning
paradigm
that
enables
models
to
be
trained
across
multiple
decentralized
devices
or
servers
holding
local
data
without
transferring
the
raw
central
location.
However,
applying
FL
heterogeneous
IoT
scenarios
comes
with
several
challenges
due
diverse
nature
of
these
in
terms
hardware
capabilities,
communications,
and
heterogeneity.
Furthermore,
conventional
parameter
server-based
aggregates
parameters
directly,
which
incurs
high
communication
overhead.
To
this
end,
paper
designs
hierarchical
federated-learning
framework
for
systems,
focusing
on
enhancing
efficiency
ensuring
security
through
lightweight
encryption.
By
leveraging
aggregation,
stream
encryption,
adaptive
device
participation,
proposed
provides
an
efficient
robust
solution
federated
learning
dynamic
resource-constrained
environments.
The
extensive
experimental
results
show
significantly
reduces
round
time
by
20%.
ACM Transactions on Autonomous and Adaptive Systems,
Journal Year:
2025,
Volume and Issue:
unknown
Published: Jan. 9, 2025
The
autonomous
driving
system
necessitates
using
privacy-preserving
deep
learning
(PPDL)
technologies
as
the
safety
assurance
for
its
extensive
application.
However,
existing
PPDL
solutions
depend
on
intricate
protocol
designs
robust
security.
Although
leveraging
advanced
dedicated
hardware
platforms
can
significantly
improve
inference
efficiency,
frameworks
that
make
best
use
of
platform
computility
are
scarce.
Thus,
balancing
efficiency
and
security
in
remains
an
open
question.
This
study
presents
SEPPDL,
a
secure
tripartite
framework
based
secret-sharing
to
balance
privacy
computational
efficiency.
We
reduce
communication
calculation
time
by
designing
quantisation
representation
scheme,
two
new
protocols,
computation
library
utilises
integer
units
GPU.
experimental
results
show
compared
with
state-of-the-art
frameworks,
SEPPDL
reduces
delay
model
1/2
1/3
optimal
while
maintaining
accuracy
inference.
Meanwhile,
achieves
10-fold
performance
improvement
lightweight
model.
As
scale
increases,
SEPPDL-based
even
86-fold
VGG16.
Mathematics,
Journal Year:
2025,
Volume and Issue:
13(5), P. 812 - 812
Published: Feb. 28, 2025
The
widespread
use
of
mobile
devices
has
led
to
the
continuous
collection
vast
amounts
user-generated
data,
supporting
data-driven
decisions
across
a
variety
fields.
However,
growing
volume
these
data
raises
significant
privacy
concerns,
especially
when
they
include
personal
information
vulnerable
misuse.
Differential
(DP)
emerged
as
prominent
solution
enabling
for
decision-making
while
protecting
user
privacy.
Despite
their
strengths,
existing
DP-based
frameworks
are
often
faced
with
trade-off
between
utility
and
computational
overhead.
To
address
challenges,
we
propose
differentially
private
fractional
coverage
model
(DPFCM),
framework
that
adaptively
balances
overhead
according
requirements
decisions.
DPFCM
introduces
two
parameters,
α
β,
which
control
fractions
collected
elements
respectively,
ensure
both
diversity
representative
coverage.
In
addition,
probability-based
methods
effectively
determining
minimum
each
should
provide
satisfy
requirements.
Experimental
results
on
real-world
datasets
validate
effectiveness
DPFCM,
demonstrating
its
high
efficiency,
applications
requiring
real-time
decision-making.
ACM Transactions on Internet of Things,
Journal Year:
2025,
Volume and Issue:
unknown
Published: March 11, 2025
The
emergence
of
the
Internet
Things
(IoT)
has
revolutionized
service
automation,
enabling
development
smart
applications.
However,
vast
interconnectivity
IoT
devices
not
only
produces
large
volumes
data
but
also
creates
multiple
potential
attack
surfaces.
While
Machine
Learning
(ML)
offers
insights
from
data,
inherent
privacy
and
security
challenges
hinder
its
effective
utilization.
Federated
(FL)
privacy-preserving
ML
for
distributed
edge
devices.
Nevertheless,
susceptibility
to
attacks
poses
a
threat
integrity
impacting
services
To
tackle
this
challenge
identify
compromised
by
like
label-flipped
paper
introduces
an
innovative
defense
mechanism
modeled
after
human
immune
system.
Analogous
‘B’
cells,
which
detect
viruses
within
body,
Reinforcement
(RL)
agent
identifies
malicious
nodes
that
participate
in
federated
learning
enabled
IoT.
Subsequently,
FL
server,
similar
‘T’
cells
Human
systems
eliminate/destroy
infected
quarantines/discards
(that
are
clients)
their
reported
parameters.
Like
work
together
defend
body
against
infections
diseases,
RL
server
defend/secure
compromised/malicious
Specifically,
with
help
Deep
(DRL),
continually
monitors
model
updates
participating
during
training
phase
find
then
isolate
or
remove
those
(i.e.,
parameters)
at
while
aggregating
parameters
global
model.
effectiveness
proposed
approach
is
demonstrated
through
experiments,
where
detects
malicious/compromised
discards
such
nodes.
We
evaluate
our
using
numerical
results
obtained
experiments
we
observe
outperforms
existing
state-of-the-art
approaches
terms
detection
rate,
error
accuracy
enhancing
Applied Sciences,
Journal Year:
2025,
Volume and Issue:
15(6), P. 3275 - 3275
Published: March 17, 2025
This
paper
proposes
a
data
security
training
framework
based
on
symmetric
projection
space
and
adversarial
training,
aimed
at
addressing
the
issues
of
privacy
leakage
computational
efficiency
encountered
by
current
protection
technologies
when
processing
sensitive
data.
By
designing
new
loss
function
combining
autoencoders
with
proposed
method
effectively
balances
model
utility.
Experimental
results
show
that,
for
financial
time-series
tasks,
using
achieves
precision
0.95,
recall
0.91,
accuracy
0.93,
significantly
outperforming
traditional
cross-entropy
loss.
In
image
yields
0.90,
mAP@50
mAP@75
0.91
respectively,
demonstrating
its
strong
advantage
in
complex
tasks.
Furthermore,
experiments
different
hardware
platforms
(Raspberry
Pi,
Jetson,
NVIDIA
3080
GPU)
that
performs
well
low-computation
devices
exhibits
significant
advantages
high-performance
GPUs,
particularly
terms
efficiency,
good
scalability
efficiency.
The
experimental
validate
superiority
Scientific Reports,
Journal Year:
2025,
Volume and Issue:
15(1)
Published: April 16, 2025
Abstract
In
the
digital
age,
privacy
preservation
is
of
paramount
importance
while
processing
health-related
sensitive
information.
This
paper
explores
integration
Federated
Learning
(FL)
and
Differential
Privacy
(DP)
for
breast
cancer
detection,
leveraging
FL’s
decentralized
architecture
to
enable
collaborative
model
training
across
healthcare
organizations
without
exposing
raw
patient
data.
To
enhance
privacy,
DP
injects
statistical
noise
into
updates
made
by
model.
mitigates
adversarial
attacks
prevents
data
leakage.
The
proposed
work
uses
Breast
Cancer
Wisconsin
Diagnostic
dataset
address
critical
challenges
such
as
heterogeneity,
privacy-accuracy
trade-offs,
computational
overhead.
From
experimental
results,
FL
combined
with
achieves
96.1%
accuracy
a
budget
ε
=
1.9,
ensuring
strong
minimal
performance
trade-offs.
comparison,
traditional
non-FL
achieved
96.0%
accuracy,
but
at
cost
requiring
centralized
storage,
which
poses
significant
risks.
These
findings
validate
feasibility
privacy-preserving
artificial
intelligence
models
in
real-world
clinical
applications,
effectively
balancing
protection
reliable
medical
predictions.