International Core Journal of Engineering 2020-26 | Page 105

intelligen*)) OR (AI artificial intellegen*) OR (Deep learning
theory) OR (capsule networks) OR (deep reinforcement
learning) OR (generative adversarial networks) OR (lean and
augmented data learning) OR (unmanned AND (vehicles OR
aerospace OR autonomous OR airborne OR (UA aircraft)))
OR (UAV aerial vehicl*) OR (Drone) OR (RPA remotely
piloted aircraft) OR (uninhabited aircraft) OR (upper
atmosphere vehicle) OR (pilotless AND ((automob*) OR
(PA airpla*))) (project selected with funding greater than or
equal to 500000).
field of artificial intelligence, and paid close attention to
research hot-spots and development trends in the field, thus
achieving scientific layout and rapid development [7]. It is
right based on above reasons that this paper takes the data of
fund sponsored projects in the field of artificial intelligence
from 2013 to 2018 as the research subjects , and the
PLDA(Parallel Latent Dirichlet Allocation)model was used
to reveals the research front in order to provide reference for
the development of artificial intelligence in our country.
II. M ETHOD
(2) (Probabilistic programming) OR (hybrid learning
models) OR (automated machine learning) OR (digital twin)
OR (explainable artificial intelligence).
A. PLDA topic model
The joint generation probability of topics and words in mth
articles is :
⃑, ⃑| ⃑, ⃑ =
Δ
⃑+ ⃑
Δ(
Δ ⃑
After removing duplicate data, a total of 9467 items of
fund projects were obtained.
⃑ + ⃑)
Δ( ⃑)
C. Research process
First, the year of 2013 is selected as the starting point of
the project, and the year is used as the unit of time for slicing,
and a total of six sub-periods are obtained. Then the
logarithmic likelihood method is used to calculate the number
of topics in each sub-period. Secondly, the PLDA model
within Knime platform is selected to identify the research
topics in each sub-period. Thirdly, the thematic modeling was
carried out for fund projects data set of the six sub-periods
from 2013 to 2018. And the multi-dimensional mapping
relations of theme-topic words and projects serial number
were obtained, so as to calculate the funding time, funding
amount and centrality of the topic-related fund projects.
Fourthly, according to the Pareto Law, the threshold of the
front identification indexes is defined, the research front topic
was determined and interpreted.
where:α and β are LDA’s prior parameters; ⃗ is the
observation data obtained by statistical counting of each word
of kth topic according to the word number 1 㸫 V. The
probability of the count is the multinomial distribution of (N,
V). N is the total number of words in kth topic. V is the total
⃗ is the observation data obtained
number of words.
from statistical counting of the topic word in the mth article
according to the article number from 1 to k. The probability of
the count is the multiple distribution of (N, K), N is the total
number of words in the mth article, K is the total number of
topics [8].
B. Research data
Artificial intelligence mainly includes six study aspects:
knowledge representation and processing, natural language
processing, machine learning, machine perception and pattern
recognition, intelligent system and application, cognition and
neuroscience inspired artificial intelligence. Because of its
wide range of contents, the retrieval scheme is based on the
method of literature research [9-15] and expert consultation.
Based on the data of artificial intelligence related fund projects
funded by NSF in the United States, the time span is from
January 1, 2013 to June 1, 2018. The retrieval is as follows:
D. Front topic identification index
Using the funding time index (FTI), the funding amount
index (FAI) and the centrality index (ECI), which are put
forward by Wang Yuyue [3] and so on, as the research front
topic identification indexes. According to the Pareto Law, the
first 20% value of the three indexes were taken as the
threshold value of front topic identification indexes, as shown
in Table I.
The decision basis is that when the three indexes of
research topic are all greater than (including equal to) the
threshold value of front topic identification indexes, it is
judged as the research front topic.
(1) (robot*) OR (neural network) OR ((speech OR image)
AND recognition) OR (machine learning) OR (computer
vision) OR (natural language processing) OR (intelligen*
system) OR (fuzzy logic) OR (brain AND (machine OR
T ABLE Ⅰ. T HRESHOLD VALUE OF FRONT TOPIC IDENTIFICATION INDEXES ACCORDING TO P ARETO L AW .
Year
2018
2017
2016
2015
2014
2013
Total number of
fund projects
39
55
38
46
43
37
Total number of top
20% projects
8
11
8
9
9
7
Funding amount
index(US$/Project)
353545
540969. 38
627436. 74
690815. 9
900555. 62
1915062. 86
Funding time
index(year/project)
4. 7
4. 07
4. 25
4. 76
5. 51
6. 43
Centrality index
7
12
11
18
9
9
an annual increasing trend. Especially in 2015, the number of
projects approved nearly doubled compared with the previous
years, which indicates that since 2015, the United States
Scientific Foundation has attached great importance to the
research and development of artificial intelligence. The
funding has increased substantially.
Since the data
acquisition deadline is June 1, 2018, the number of projects
approved in 2018 is small.
III. R ESULTS
A. Annual distribution of the project number funded by NSF
in artificial intelligence field
From 2013 to 2018 years, the number of projects
approved by NSF in the field of artificial intelligence in the
United States was 642, 1318, 2110, 2343, 2469, 584, showing
83