Nature Communications,
Год журнала:
2021,
Номер
12(1)
Опубликована: Ноя. 15, 2021
Abstract
Artificial
intelligence
(AI)
and
machine
learning
(ML)
have
been
increasingly
used
in
materials
science
to
build
predictive
models
accelerate
discovery.
For
selected
properties,
availability
of
large
databases
has
also
facilitated
application
deep
(DL)
transfer
(TL).
However,
unavailability
datasets
for
a
majority
properties
prohibits
widespread
DL/TL.
We
present
cross-property
deep-transfer-learning
framework
that
leverages
trained
on
small
different
properties.
test
the
proposed
39
computational
two
experimental
find
TL
with
only
elemental
fractions
as
input
outperform
ML/DL
from
scratch
even
when
they
are
allowed
use
physical
attributes
input,
27/39
(≈
69%)
both
datasets.
believe
can
be
widely
useful
tackle
data
challenge
applying
AI/ML
science.
The Journal of Physical Chemistry Letters,
Год журнала:
2025,
Номер
unknown, С. 518 - 524
Опубликована: Янв. 6, 2025
Evaluating
the
quantum
optical
properties
of
solid-state
single-photon
emitters
is
a
time-consuming
task
that
typically
requires
interferometric
photon
correlation
experiments.
Photon
Fourier
spectroscopy
(PCFS)
one
such
technique
measures
time-resolved
single-emitter
line
shapes
and
offers
additional
spectral
information
over
Hong–Ou–Mandel
two-photon
interference
but
long
experimental
acquisition
times.
Here,
we
demonstrate
neural
ordinary
differential
equation
model,
g2NODE,
can
forecast
complete
noise-free
interferometry
experiment
from
small
subset
noisy
functions.
We
this
for
simulated
data,
where
g2NODE
utilizes
10–20
measured
functions
to
create
entire
denoised
interferograms
up
200
stage
positions,
enabling
20-fold
speedup
in
time
hours
minutes.
Our
work
presents
new
deep
learning
approach
greatly
accelerate
use
as
an
characterization
tool
novel
emitter
materials.
Chemical Science,
Год журнала:
2025,
Номер
unknown
Опубликована: Янв. 1, 2025
MOSAEC-DB
represents
the
largest
and
most
diverse
dataset
of
experimental
MOFs
suitable
for
simulation
machine
learning
applications.
Novel
approaches
utilizing
metal
oxidation
states
enhance
its
chemical
accuracy
relative
to
past
MOF
databases.
Nature Communications,
Год журнала:
2021,
Номер
12(1)
Опубликована: Ноя. 15, 2021
Abstract
Artificial
intelligence
(AI)
and
machine
learning
(ML)
have
been
increasingly
used
in
materials
science
to
build
predictive
models
accelerate
discovery.
For
selected
properties,
availability
of
large
databases
has
also
facilitated
application
deep
(DL)
transfer
(TL).
However,
unavailability
datasets
for
a
majority
properties
prohibits
widespread
DL/TL.
We
present
cross-property
deep-transfer-learning
framework
that
leverages
trained
on
small
different
properties.
test
the
proposed
39
computational
two
experimental
find
TL
with
only
elemental
fractions
as
input
outperform
ML/DL
from
scratch
even
when
they
are
allowed
use
physical
attributes
input,
27/39
(≈
69%)
both
datasets.
believe
can
be
widely
useful
tackle
data
challenge
applying
AI/ML
science.
