INFORMS journal on computing,
Год журнала:
2024,
Номер
unknown
Опубликована: Июнь 6, 2024
We
investigate
a
novel
type
of
online
sequential
decision
problem
under
uncertainty,
namely
mixed
observability
Markov
process
with
time-varying
interval-valued
parameters
(MOMDP-TVIVP).
Such
data-driven
optimization
problems
learning
widely
have
real-world
applications
(e.g.,
coordinating
surveillance
and
intervention
activities
limited
resources
for
pandemic
control).
Solving
MOMDP-TVIVP
is
great
challenge
as
system
identification
reoptimization
based
on
newly
observational
data
are
required
considering
the
unobserved
states
parameters.
Moreover,
many
practical
problems,
action
state
spaces
intractably
large
optimization.
To
address
this
challenge,
we
propose
transfer
reinforcement
(TRL)-based
algorithmic
approach
that
ingrates
(TL)
into
deep
(DRL)
in
an
offline-online
scheme.
accelerate
reoptimization,
pretrain
collection
promising
networks
fine-tune
them
acquired
system.
The
hallmark
our
comes
from
combining
strong
approximation
ability
neural
high
flexibility
TL
through
efficiently
adapting
previously
learned
policy
to
changes
dynamics.
Computational
study
different
uncertainty
configurations
scales
shows
outperforms
existing
methods
solution
optimality,
robustness,
efficiency,
scalability.
also
demonstrate
value
fine-tuning
by
comparing
TRL
DRL,
which
at
least
21%
improvement
can
be
yielded
no
more
than
0.62%
time
spent
pretraining
each
period
instances
continuous
state-action
space
modest
dimensionality.
A
retrospective
control
use
case
Shanghai,
China
improved
making
via
several
public
health
metrics.
Our
first-ever
endeavor
employing
intensive
network
training
solving
processes
requiring
reoptimization.
History:
Accepted
Paul
Brooks,
Area
Editor
Applications
Biology,
Medicine,
&
Healthcare.
Funding:
This
work
was
supported
part
National
Natural
Science
Foundation
[Grants
72371051
72201047]
first
second
authors
[Grant
1825725]
third
author.
Supplemental
Material:
software
supports
findings
available
within
paper
its
Information
(
https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2022.0236
)
well
IJOC
GitHub
repository
https://github.com/INFORMSJoC/2022.0236
).
complete
Software
Data
Repository
https://informsjoc.github.io/
.
European Journal of Operational Research,
Год журнала:
2024,
Номер
318(3), С. 927 - 953
Опубликована: Июнь 22, 2024
A
burgeoning
literature
shows
that
self-learning
algorithms
may,
under
some
conditions,
reach
seemingly-collusive
outcomes:
after
repeated
interaction,
competing
earn
supra-competitive
profits,
at
the
expense
of
efficiency
and
consumer
welfare.
This
paper
offers
evidence
such
behavior
can
stem
from
insufficient
exploration
during
learning
process
algorithmic
sophistication
might
increase
competition.
In
particular,
we
show
allowing
for
more
thorough
does
lead
otherwise
Q-learning
to
play
competitively.
We
first
provide
a
theoretical
illustration
this
phenomenon
by
analyzing
competition
between
two
stylized
in
Prisoner's
Dilemma
framework.
Second,
via
simulations,
sophisticated
exploit
ones.
Following
these
results,
argue
advancement
computational
capabilities
situations,
solution
challenge
seeming
collusion,
rather
than
exacerbate
it.
Journal of Global Optimization,
Год журнала:
2024,
Номер
89(3), С. 655 - 685
Опубликована: Фев. 15, 2024
Abstract
In
this
paper,
we
address
the
difficulty
of
solving
large-scale
multi-dimensional
knapsack
instances
(MKP),
presenting
a
novel
deep
reinforcement
learning
(DRL)
framework.
DRL
framework,
train
different
agents
compatible
with
discrete
action
space
for
sequential
decision-making
while
still
satisfying
any
resource
constraint
MKP.
This
framework
incorporates
decision
variable
values
in
2D
where
agent
is
responsible
assigning
value
1
or
0
to
each
variables.
To
best
our
knowledge,
first
model
its
kind
which
environment
formulated,
and
an
element
solution
matrix
represents
item
Our
configured
solve
MKP
dimensions
distributions.
We
propose
K-means
approach
obtain
initial
feasible
that
used
agent.
four
present
results
comparing
them
CPLEX
commercial
solver.
The
show
can
learn
generalize
over
sizes
shows
it
medium-sized
at
least
45
times
faster
CPU
time
10
large
instances,
maximum
gap
0.28%
compared
performance
CPLEX.
Furthermore,
95%
items
are
predicted
line
solution.
Computations
also
provide
better
optimality
respect
state-of-the-art
approaches.
Advances in human resources management and organizational development book series,
Год журнала:
2025,
Номер
unknown, С. 109 - 132
Опубликована: Фев. 7, 2025
Although
artificial
intelligence
has
driven
digital
transformation
in
several
countries
and
sectors,
many
large
companies
are
still
lagging
behind
adopting
these
technologies.
In
fact,
business
managers
remain
unaware
of
the
strategic
role
AI
can
play.
Therefore,
explaining
potential
its
implications
could
be
a
viable
solution
to
address
this
issue.
context,
chapter
explores
how
enhance
organizational
performance
by
developing
dynamic
capabilities.
Using
survey-based
approach,
we
collected
data
from
multinational
firms
Tunisia
examine
indirect
effect
adoption
on
performance.
Data
was
gathered
226
analyzed
through
structural
equation
modeling.
Our
findings
reveal
that
positively
impacts
three
key
capabilities:
exploration
innovation,
decision-making
speed,
exploitation
innovation.
These
results
highlight
benefits
firms,
fostering
capabilities
that,
turn,
Advances in computational intelligence and robotics book series,
Год журнала:
2025,
Номер
unknown, С. 427 - 454
Опубликована: Янв. 10, 2025
Machine
learning
(ML)
optimization
techniques
serve
as
essential
for
training
models
to
achieve
high
performance
in
a
diverse
areas.
This
chapter
offers
thorough
summary
of
machine
techniques.
analysis
the
development
over
time.
A
number
common
constraints
are
also
discussed.
Developing
model
that
works
effectively
and
provides
accurate
predictions
certain
set
instances
is
main
objective
ML.
We
require
ML
accomplish
that.
The
practice
modifying
hyper
parameters
with
an
technique
minimize
cost
function
called
optimization.
Because
indicates
difference
between
actual
value
estimated
parameter
predicted
by
model,
it
crucial
reduce
it.
will
provide
general
explanation
workings
drawbacks
strategies.
Numerous
advancements
have
been
put
forth
this
chapter.
ABSTRACT
In
recent
years,
artificial
intelligence
(AI)
has
made
significant
strides
in
research
and
shown
great
potential
various
application
fields,
including
business
economics
(B&E).
However,
AI
models
are
often
black
boxes,
making
them
difficult
to
understand
explain.
This
challenge
can
be
addressed
using
eXplainable
Artificial
Intelligence
(XAI),
which
helps
humans
the
factors
driving
decisions,
thereby
increasing
transparency
confidence
results.
paper
aims
provide
a
comprehensive
understanding
of
state‐of‐the‐art
on
XAI
B&E
by
conducting
an
extensive
literature
review.
It
introduces
novel
approach
categorising
techniques
from
three
different
perspectives:
samples,
features
modelling
method.
Additionally,
identifies
key
challenges
corresponding
opportunities
field.
We
hope
that
this
work
will
promote
adoption
B&E,
inspire
interdisciplinary
collaboration,
foster
innovation
growth
ensure
explainability.
