A Survey of Intelligent End-to-End Networking Solutions: Integrating Graph Neural Networks and Deep Reinforcement Learning Approaches
Electronics,
Год журнала:
2024,
Номер
13(5), С. 994 - 994
Опубликована: Март 6, 2024
This
paper
provides
a
comprehensive
survey
of
the
integration
graph
neural
networks
(GNN)
and
deep
reinforcement
learning
(DRL)
in
end-to-end
(E2E)
networking
solutions.
We
delve
into
fundamentals
GNN,
its
variants,
state-of-the-art
applications
communication
networking,
which
reveal
potential
to
revolutionize
access,
transport,
core
network
management
policies.
further
explores
DRL
capabilities,
trending
E2E
particularly
enhancing
dynamic
(re)configurations
resource
management.
By
fusing
GNN
with
DRL,
we
spotlight
novel
approaches,
ranging
from
radio
access
orchestration,
across
layers.
Deployment
scenarios
smart
transportation,
factory,
grids
demonstrate
practical
implications
our
topic.
Lastly,
point
out
challenges
future
research
directions,
including
critical
aspects
for
modelling
explainability,
reduction
overhead
consumption,
interoperability
existing
schemes,
importance
reproducibility.
Our
aims
serve
as
roadmap
developments
guiding
through
current
landscape,
challenges,
prospective
breakthroughs
algorithm
toward
automation
using
DRL.
Язык: Английский
Handling Efficient VNF Placement with Graph-Based Reinforcement Learning for SFC Fault Tolerance
Electronics,
Год журнала:
2024,
Номер
13(13), С. 2552 - 2552
Опубликована: Июнь 28, 2024
Network
functions
virtualization
(NFV)
has
become
the
platform
for
decomposing
sequence
of
virtual
network
(VNFs),
which
can
be
grouped
as
a
forwarding
graph
service
function
chaining
(SFC)
to
serve
multi-service
slice
requirements.
NFV-enabled
SFC
consists
several
challenges
in
reaching
reliability
and
efficiency
key
performance
indicators
(KPIs)
management
orchestration
(MANO)
decision-making
control.
The
problem
fault
tolerance
is
one
most
critical
provisioning
requests,
it
needs
resource
availability.
In
this
article,
we
proposed
neural
(GNN)-based
deep
reinforcement
learning
(DRL)
enhance
(GRL-SFT),
targets
chain
representation,
long-term
approximation,
self-organizing
future
massive
Internet
Everything
applications.
We
formulate
Markov
decision
process
(MDP).
DRL
seeks
maximize
cumulative
rewards
by
maximizing
request
acceptance
ratios
minimizing
average
completion
delays.
model
solves
VNF
short
time
configures
node
allocation
reliably
real-time
restoration.
Our
simulation
result
demonstrates
effectiveness
scheme
indicates
better
terms
total
rewards,
delays,
acceptances,
failures,
restoration
different
topologies
compared
reference
schemes.
Язык: Английский
Efficient Fabric Classification and Object Detection Using YOLOv10
Electronics,
Год журнала:
2024,
Номер
13(19), С. 3840 - 3840
Опубликована: Сен. 28, 2024
The
YOLO
(You
Only
Look
Once)
series
is
renowned
for
its
real-time
object
detection
capabilities
in
images
and
videos.
It
highly
relevant
industries
like
textiles,
where
speed
accuracy
are
critical.
In
the
textile
industry,
accurate
fabric
type
classification
essential
improving
quality
control,
optimizing
inventory
management,
enhancing
customer
satisfaction.
This
paper
proposes
a
new
approach
using
YOLOv10
model,
which
offers
enhanced
accuracy,
processing
speed,
on
torn
path
of
each
fabric.
We
developed
utilized
specialized,
annotated
dataset
featuring
diverse
samples,
including
cotton,
hanbok,
cotton
yarn-dyed,
blend
plain
fabrics,
to
detect
model
was
selected
superior
performance,
leveraging
advancements
deep
learning
architecture
applying
data
augmentation
techniques
improve
adaptability
generalization
various
patterns
textures.
Through
comprehensive
experiments,
we
demonstrate
effectiveness
YOLOv10,
achieved
an
85.6%
outperformed
previous
variants
both
precision
speed.
Specifically,
showed
2.4%
improvement
over
YOLOv9,
1.8%
YOLOv8,
6.8%
YOLOv7,
5.6%
YOLOv6,
6.2%
YOLOv5.
These
results
underscore
significant
potential
automating
processes,
thereby
operational
efficiency
productivity
manufacturing
retail.
Язык: Английский
Dynamic Bandwidth Slicing in Passive Optical Networks to Empower Federated Learning
Sensors,
Год журнала:
2024,
Номер
24(15), С. 5000 - 5000
Опубликована: Авг. 2, 2024
Federated
Learning
(FL)
is
a
decentralized
machine
learning
method
in
which
individual
devices
compute
local
models
based
on
their
data.
In
FL,
periodically
share
newly
trained
updates
with
the
central
server,
rather
than
submitting
raw
The
key
characteristics
of
including
on-device
training
and
aggregation,
make
it
interesting
for
many
communication
domains.
Moreover,
potential
new
systems
facilitating
FL
sixth
generation
(6G)
enabled
Passive
Optical
Networks
(PON),
presents
promising
opportunity
integration
within
this
domain.
This
article
focuses
interaction
between
PON,
exploring
approaches
effective
bandwidth
management,
particularly
addressing
complexity
introduced
by
traffic.
PON
standard,
advanced
management
proposed
allocating
multiple
upstream
grants
utilizing
Dynamic
Bandwidth
Allocation
(DBA)
algorithm
to
be
allocated
an
Network
Unit
(ONU).
However,
there
lack
research
studying
utilization
grant
allocation.
paper,
we
address
limitation
introducing
novel
DBA
approach
that
efficiently
allocates
traffic
demonstrates
how
can
benefit
from
enhanced
capacity
implementing
carrying
out
flows.
Simulations
conducted
study
show
solution
outperforms
state-of-the-art
solutions
several
network
performance
metrics,
reducing
delay.
improvement
holds
great
promise
enabling
real-time
data-intensive
services
will
components
6G
environments.
Furthermore,
our
discussion
outlines
as
operational
reality
capable
supporting
networking.
Язык: Английский
Priority/Demand-Based Resource Management with Intelligent O-RAN for Energy-Aware Industrial Internet of Things
Processes,
Год журнала:
2024,
Номер
12(12), С. 2674 - 2674
Опубликована: Ноя. 27, 2024
The
last
decade
has
witnessed
the
explosive
growth
of
internet
things
(IoT),
demonstrating
utilization
ubiquitous
sensing
and
computation
services.
Hence,
industrial
IoT
(IIoT)
is
integrated
into
devices.
IIoT
concerned
with
limitation
battery
life.
Therefore,
mobile
edge
computing
(MEC)
a
paradigm
that
enables
proliferation
resource
reduces
network
communication
latency
to
realize
perspective.
Furthermore,
an
open
radio
access
(O-RAN)
new
architecture
adopts
MEC
server
offer
provisioning
framework
address
energy
efficiency
reduce
congestion
window
IIoT.
However,
dynamic
continuity
task
generation
by
lead
challenges
in
management
orchestration
(MANO)
efficiency.
In
this
article,
we
aim
investigate
priority
on
demand.
Additionally,
minimize
long-term
average
delay
resource-intensive
tasks,
Markov
decision
problem
(MDP)
conducted
solve
problem.
deep
reinforcement
learning
(DRL)
optimal
handling
policy
for
MEC-enabled
O-RAN
architectures.
study,
MDP-assisted
q-network-based
priority/demanding
management,
namely
DQG-PD,
been
investigated
optimizing
management.
DQG-PD
algorithm
aims
devices,
which
demonstrates
exploiting
Q-network
(DQN)
jointly
optimizes
each
service
request.
DQN
divided
online
target
networks
better
adapt
environment.
Finally,
our
experiment
shows
work
can
outperform
reference
schemes
terms
resources,
cost,
energy,
reliability,
completion
ratio.
Язык: Английский