World Journal of Innovation and Modern Technology, Vol.

7, Issue 2,
1, (Apr)
(Feb)
ISSN 2682-5910 2024
2023
DOI: 10.53469/wjimt.2024.07(03).03

Current State of Autonomous Driving

Applications Based on Distributed Perception
and Decision-Making
Baoming Wang1,*, Han Lei2, Zuwei Shui3, Zhou Chen4, Peiyuan Yang5
1
Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, USA
2
Computer Science Engineering, Santa Clara University, Santa Clara, USA
3
Information Studies, Trine University, Phoenix, USA
4
Software Engineering, ZhejiangUniversity, Hangzhou, China
5
Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, USA
*Corresponding Author, [email protected]

Abstract: This article reviews the key role of distributed cloud architecture in autonomous driving systems and its
integration with intelligent computing networks. By spreading computing resources across multiple geographic locations,
the distributed cloud enables localized processing and storage of data, reducing latency and improving real-time decision
making in autonomous vehicles. The article points out that the combination of distributed cloud technology and intelligent
computing network provides a powerful solution to meet the challenges of autonomous driving technology. By dynamically
allocating computing resources and deeply integrating cloud, network, and chip technologies, distributed cloud gives
autonomous driving systems enhanced data processing capabilities to ensure stable and reliable performance in a variety of
driving scenarios. Finally, the paper highlights that the synergy of distributed cloud and intelligent driving technology
marks an important milestone for intelligent transportation systems, heralding the accelerated adoption of distributed cloud
solutions in the automotive industry, driving the pace of innovation and transformation.

Keywords: Distributed cloud architecture; Autonomous driving system; Data localization processing; Intelligent perception
and decision making; Real-time decision capability

1. INTRODUCTION
As an essential facet of future transportation, autonomous driving technology is encountering unparalleled
challenges in computational power. Autonomous vehicles must adeptly discern intricate environmental cues such
as pedestrians, vehicles, and traffic signs while navigating at high speeds, necessitating real-time decision-making
and control. This mandates that the autonomous driving system possess robust data processing capabilities and
computational power support. [1-3]According to authoritative forecasts, the global self-driving car market is
poised to exceed $100 billion by 2025. Industry statistics reveal a staggering surge in the requisite computational
power for self-driving cars. An advanced self-driving vehicle may necessitate computing power equivalent to
thousands of conventional servers combined.Confronted with such immense computational demands, the
traditional cloud computing model struggles to suffice. Traditionally, this model centralizes data storage in remote
data centers for processing, introducing challenges of data transmission latency and potential risks to data security
and privacy protection.

2. RELATED WORK
2.1 Distributed cloud

Distributed cloud is different from distributed computing, and the concept of distributed cloud is still from the
2021 strategic technology trend forecast released by Garnterat the end of 2020, one of the most important concepts
is distributed cloud, which is Gartner's second time to mention distributed cloud as a strategic technology. Let's
take a look at how Garnter defines a distributed cloud.

Distributed cloud: [4]The distribution of public cloud services (often including the necessary hardware and
software) to different physical locations (i.e., the edge), while the ownership, operation, governance, updating, and
development of the service remains the responsibility of the original public cloud provider.

15
 2024 World Journal of Innovation and Moder
ISSN 2682-5910
ernn Tec
echhnology, Vol.7, Issue 3 (Jun)

Gartner also predicts that by 2025, more than 50 percent of organizations will use distributed cloud in their chosen
locations, enabling a transformational business model. [5]In literal terms, distributed cloud refers to the
redistribution of public cloud services in physical locations, but operations remain unified. Our careful analysis
shows that this is actually consistent with the premise of the multi-cloud concept. The distributed cloud embodies
the principle of proximity of computing power, which in fact solves the needs of customers to keep cloud
computing resources close to the physical location of data and business activities.

Distributed cloud has two key points in the architecture, one is distributed or decentralized, and the other is
centralized or unified.[6]Decentralized is the bearer of different business attributes and regional requirements,
centralized is the management and standards.

Distributed cloud Features:

1. High availability

A distributed cloud architecture avoids the risk of a single point of failure by spreading resources across multiple
locations. [7]Even if one node fails, other nodes can still provide services to ensure business continuity and
reliability.

2. Scalability

The distributed cloud can flexibly adjust the scale of resources based on service requirements. When the service
load increases, you can add more nodes to expand the computing and storage capabilities to meet user
requirements. On the contrary, when the service load is reduced, the number of nodes can be appropriately reduced
to save resources and costs.

3. Flexibility and flexibility

The distributed cloud is characterized by elasticity and flexibility. It can automatically adjust resources on demand
to accommodate traffic peaks and valleys. [8]This flexibility enables the distributed cloud to cope with sudden
changes in business demands and traffic, providing stable performance and user experience.

4. Data locality

The distributed cloud allows data to be stored on the node closest to the user, reducing latency and bandwidth
consumption in data transmission. This advantage of data locality can improve the efficiency and responsiveness
of data access, especially for application scenarios and cases distributed in different regions of the world

2.2 Distributed systems and autonomous driving

With the continuous improvement of the performance of AI chips and sensors such as lidar, millimeter-wave radar,
and cameras, coupled with the gradual reduction of costs, the line and mooring integrated program gradually began
to be mass-produced. In the future, there will be a large number of new cars equipped with an integrated solution
on the road, which can collect more valuable data, and then promote the continuous improvement and evolution of
high-level intelligent driving functions such as NOA and AVP through a data-driven way. The distributed cloud
provides an entirely new solution. [9-11]By decentralizing cloud computing resources across multiple geographic
locations, the distributed cloud enables localized processing and storage of data. According to industry experts,
through the distributed cloud architecture, the autonomous driving system can achieve localized processing,
reduce data transmission delay, and improve real-time. [12]Taking Kehua Data's solution as an example, its
distributed intelligence center can reduce data processing latency to the millisecond level, ensuring that the
autonomous driving system can make rapid and accurate decisions in a variety of complex environments.

In addition, the distributed cloud also has high reliability and high scalability. By building multiple redundant data
centers and computing nodes, the distributed cloud can ensure stable and reliable computing support for
autonomous driving systems in any situation. At the same time, with the continuous expansion of the autonomous
driving market and the continuous progress of technology, distributed cloud can also be flexibly expanded and
upgraded according to demand.

16
 World Journal of Innovation and Modern Technology, Vol. 7, Issue 2,
1, (Apr)
(Feb)
ISSN 2682-5910 2024
2023

By dynamically allocating computing resources, the computing power network can meet the requirements of the
autonomous driving system in different scenarios. Through the deep integration of cloud, network and chip
technology, Kehua Data realizes the efficient construction of computing power network, and provides strong
computing power support for the autonomous driving industry. [13]Taking Kehua Data's distributed intelligent
computing center as an example, its computing power network has been successfully applied to a number of
autonomous driving projects. [14-16]According to practical application cases, the computing power network can
increase the processing speed of the autonomous driving system several times, while reducing power consumption
and cost. This successful case not only proves the practical application value of distributed cloud and computing
power network in the autonomous driving industry, but also provides valuable experience and reference for other
enterprises.

In summary, distributed cloud and computing power networks have become the biggest demand of the
autonomous driving industry. By providing powerful data processing and computing support, they provide a solid
guarantee for the innovative development and application of autonomous driving technology. In the future, with
the further popularization and application of autonomous driving technology, the demand for distributed cloud and
computing power networks will be more vigorous. [17][18]We look forward to more companies joining this field
and jointly promoting the prosperity of the autonomous driving industry.

2.3 Autonomous driving perception and decision making

Figure 1 below gives a clear picture of how an autonomous vehicle system is structured. In this diagram, you'll see
two main parts: the perception system and the decision system, each represented by different colored modules.

The perception system is like the eyes and ears of the autonomous vehicle. It gathers information from various
sensors onboard the vehicle, such as [19]LIDAR, radar, cameras, [20]GPS, [21]IMUs, and odometers. These
sensors help the vehicle understand its surroundings, including detecting objects, understanding road conditions,
and keeping track of its own position and movement. The perception system also relies on prior knowledge about
sensor capabilities, road networks, traffic rules, and vehicle behavior.

On the other hand, the decision system is like the brain of the autonomous vehicle. It takes the information
provided by the perception system and uses it to make decisions about how the vehicle should navigate through its
environment. [22]This includes tasks like planning the vehicle's route, avoiding obstacles, obeying traffic laws,
and ensuring passenger safety and comfort.

Overall, this architectural block diagram illustrates how an autonomous vehicle combines perception and
decision-making capabilities to navigate safely and efficiently in its environment.

Figure 1. Overall operational framework of automatic driving

17
 2024 World Journal of Innovation and Moder
ISSN 2682-5910
ernn Tec
echhnology, Vol.7, Issue 3 (Jun)

The decision system guides the car from its starting point to a destination defined by the user. It considers the car's
condition, the environment, road rules, and passenger comfort. [23]To navigate effectively, it needs to know where
the car is in its surroundings. This is where the locator module comes in. It estimates the car's position, speed, and
direction using a map of the environment. This map is created before the car operates autonomously, often using its
own sensors. Sometimes, manual adjustments are needed, like marking crosswalks or removing non-static objects.
The positioning module takes offline maps, sensor data, and odometer readings as inputs and outputs the car's
status. [24]While GPS can assist, it's not always reliable in urban areas due to interference. The mapper module
then uses offline maps and current data to create online maps. These online maps combine static map information
with real-time sensor data. They're crucial for decision-making and can include a Moving Object tracking module
to detect and remove dynamic elements.

In this part, we discuss key methods for autonomous vehicle perception systems, like locating the vehicle, mapping
obstacles, roads, tracking moving obstacles, and recognizing traffic signals. The positioning module estimates the
car's position and direction relative to the map or road, often using GPS[25]. However, GPS isn't always reliable in
urban areas. Other methods, like [26]LiDAR, LiDAR plus cameras, or just cameras, offer alternatives. LiDAR
provides accurate measurements but can be costly. Combining LiDAR with cameras reduces costs while
maintaining accuracy. Camera-based methods are cheaper but may be less precise.

3. METHODOLOGY
3.1 Experimental design

Data closed loop plays a crucial role in the development of autonomous driving technology. Understanding its
composition is critical to advancing the development of autonomous driving technology. The following are the
main components of a data closed loop:

1. Data collection: Data collection is carried out by professional collection vehicles, test vehicles or production
vehicles. These vehicles are equipped with a variety of sensors and data recording devices to capture various
scenarios and situations in the real world[27].

2. Data return: The collected data is processed locally, including classification, desensitization, compression and
packaging, and then uploaded to the cloud server through 4G/5G and other networks. This ensures the safe
transmission and storage of data.

3. Data annotation: Annotate the collected data on the cloud server. Annotation refers to giving semantic
information to data, such as identifying vehicles, pedestrians, traffic lights, etc. [28]This process is crucial for
training and validating autonomous driving systems.

4. Model training: Marked data is sent to the training platform for training machine learning models. These models
include parts such as perception, decision-making, and are used for all aspects of autonomous driving systems.

5. Simulation test: After model training, simulation test is required. Simulation testing is to simulate real world
situations through virtual scenarios to verify and evaluate the performance and safety of the trained model.

By closing the loop on these key links, autonomous driving technology can continuously collect data from the real
world, train models, and optimize and improve its performance through simulation tests. This circular feedback
process is an important guarantee for the realization of high-level intelligent driving perception and
decision-making.

3.2 Data acquisition

After the equivalent production vehicle reaches a certain scale, the amount of data collected will be large. Some
professionals estimate that if the total amount of data collected in accordance with 100,000 vehicles is estimated
for 300 days a year, the total amount of data faced by car companies in the future will reach ZB level. If so much
data is uploaded to the cloud without effective screening, it will also bring great pressure on transmission
bandwidth, data storage and processing in the cloud. [29]Therefore, how to select valuable data from massive data
for efficient return is the key to affecting the iteration speed of high-level intelligent driving system.

18
 World Journal of Innovation and Modern Technology, Vol. 7, Issue 2,
1, (Apr)
(Feb)
ISSN 2682-5910 2024
2023

There are several common intelligent data acquisition methods:

A. Set the Trigger layer at the car end

According to model failure analysis and model decision boundary analysis, the scene to be collected is set in
advance and the collection logic is formulated. Then, the trigger layer (data return trigger) is set at the end of the
vehicle, and the required scene data set is automatically obtained according to the scene algorithm detection.

B. Intelligently label data for specific scenarios

For some scenarios that need to actively accumulate data in the cloud for learning, such as complex working
conditions such as tunnels, roundabout islands, unprotected left turns, etc., developers can upload pictures that
need to be obtained by the vehicle, issue instructions through the cloud, and the vehicle will take a similar way of
"image search" to automatically intercept similar scenes.

3.3 Data Sending

After collecting the data, the data needs to be classified, desensitized, compressed and packaged, and uploaded to
the cloud server through 4G/5G[30]. However, the uplink and downlink of data transmission is relatively long, and
the link of the car network is usually not stable, and the car may cross different base stations in the process of fast
driving, there is a 4G-5G switch, which will lead to the loss and disorder of the signal in the transmission process.
So, how to ensure the data transmission speed and quality of each link node?

In view of some problems in data transmission, some enterprises have adopted a vehicle-cloud integrated
transmission scheme. For example, Zhixhuitong combines data slices through SDK vCloud of cloud data
management, adopts a verification and retransmission mechanism for slices, retransmits lost data slices, and stores
them after data retransmission is complete, ensuring the integrity of data transmission.

3.4 Data Annotation

After the car end data is returned to the cloud, it also needs to be marked with data to become sample data that can
be used by the algorithm model. [31-32]Labeling is the process of assigning coded values to raw data, thus using
the labeled data as training data for AI to practice cognition. Encoded values include, but are not limited to,
assigning class labels, drawing bounding boxes, and marking object boundaries.

With the continuous iteration of intelligent driving technology, intelligent driving system has higher and higher
requirements for the accuracy of perception model. Therefore, to improve the accuracy of vehicle perception
models, large-scale and high-quality data sets are needed for training. Traditional manual annotation has been
unable to meet the demand for massive data sets in model training in terms of efficiency and cost, and a new
annotation method is needed to improve the annotation quality and efficiency.

The pre-labeling algorithm can greatly reduce the time required for each box of data to be labeled. Baidu
autonomous driving cloud technology experts have said: "Before the annotation, we will first use the algorithm to
do a pre-annotation, which can greatly improve the efficiency of the tagger single frame annotation." In the tagging
process, we introduce a lot of intelligent algorithms to assist our taggers. For example, when doing area
segmentation, we will learn from the edging algorithm similar to photoshop to achieve better fitting effect and
improve the tagging efficiency."

3.5 Model Training

After the completion of data annotation, it is necessary to train the labeled data and adjust parameters. The training
of large models used in advanced intelligent driving has a high demand for computing power. Some car companies
have specially created their own intelligent computing centers, such as Tesla's Dojo, Geely's Star Rui intelligent
computing Center, Xiaopeng's "Fuyao", and the Snow Lake · OASIS (MANA OASIS).

Model training is to explore and analyze data through analytical means and methods to find causality, data logic
and business laws. In the model training, we can design a set of automatic training engine with the help of Auto ML
and other tools to automate part of the model training.

19
 2024 World Journal of Innovation and Moder
ISSN 2682-5910
ernn Tec
echhnology, Vol.7, Issue 3 (Jun)

In addition, from the perspective of data operation, the efficiency and quality of model training can also be
improved. Baidu has taken the following measures:

Help customers build the most effective data set in the model training process - help customers to plan what data
needs to be marked, what categories to mark, and what its distribution is in the early stage.
For the marked data, Baidu Intelligent Cloud will help enterprises evaluate their models according to the existing
huge evaluation sample set, and discover the bad cases or shortcomings of the current model. In view of the
shortcomings of the current model, it is necessary to supplement enough training sets to help the enterprise
improve its model indicators and optimize the model.

3.6 Requirements for cloud access for perceptual model training

The training of automatic driving perception model requires large data volume, high algorithm accuracy and
training efficiency, which is suitable for cloud, and adopts cloud service for data processing. CNN is a common
deep learning model in intelligent driving perception, and its design is basically aimed at INT8. BF16 is the most
suitable format for Transformer model architecture. Some industry insiders admitted: "Transformer belongs to
violent aesthetics, compared with CNN, its model is larger, the number of participants is one billion billion, one
billion trillion is not rare, the number of layers is thousands of layers, it is not the old data training center can
support."

Among them, for L2 and L3 levels, if you do L2 and L3 level Demo, you only need millions of pictures and several
GPU cards to train. Due to the small amount of data and limited scope of use, the requirements are not very high
from the perspective of compliance, and it probably needs several million dollars of investment. If you do L2 and
L3 level mass production, you need hundreds of millions of pictures and 100+ GPU cards for training scheduling,
and the capital investment can reach tens of millions.

For L4 level, if you do L4 level Demo, the amount of data required to store more than 1PB, 100+ GPU cards are
required for training scheduling, and the investment scale reaches 50 million to 200 million. If you want to achieve
L4 level mass production, the amount of data required to store will exceed 50PB, 500+ GPU cards are required for
training scheduling, and the related investment will exceed 500 million levels.

3.7 Distributed Cloud Simulation

The traditional single-machine simulation test presents some problems, such as insufficient computing power and
unable to realize accelerated testing, such as long test cycle and low efficiency. Cloud platform simulation, with its
distributed architecture and parallel accelerated computing capabilities, can greatly improve the efficiency of
simulation testing, and is an effective solution to achieve large-scale simulation scenarios for autonomous driving.
Although cloud simulation can effectively improve the efficiency of simulation test, the industry still has the
following questions about simulation test: whether the simulation scene is true, the simulation scene is incomplete,
the iteration speed is not fast, the simulation evaluation is accurate or not.

For these problems, Baidu Cloud simulation platform has carried out targeted improvements and solutions:

In terms of the reality of the scene, in accordance with compliance standards, the integration of high-precision
maps, 1:1 depiction of the physical world of the road topology. For dynamic traffic participation elements, mining
is done based on real road data, and the interaction of dynamic elements is accurately described.

In the scene generation mode, the combination of manual scene editing mode and scene mining based on real road
data has covered 98% of the scenes (including cities, highways, parking lots, closed parks, etc.), according to
Baidu insiders.

In terms of iteration speed, relying on the technical support and computing power advantages of Baidu Intelligent
Cloud, Baidu's cloud simulation platform can realize the concurrent operation of hundreds of thousands of tasks,
and achieve a daily simulation mileage of tens of thousands of kilometers.
In terms of simulation evaluation standards, Baidu has summarized more than 200 evaluation indicators in six
categories after several years of experience. In addition to safety and traffic regulations, comfort and intelligence
are also added to the evaluation criteria through rules.

20
 World Journal of Innovation and Modern Technology, Vol. 7, Issue 2,
1, (Apr)
(Feb)
ISSN 2682-5910 2024
2023

4. CONCLUSION
Based on the comprehensive analysis of distributed cloud architecture and its application in autonomous driving
systems, it is evident that distributed cloud technology plays a pivotal role in enhancing the efficiency, reliability,
and scalability of autonomous driving functions. By decentralizing computing resources across multiple
geographic locations, distributed cloud enables localized processing and storage of data, thereby reducing latency
and improving real-time decision-making capabilities in autonomous vehicles.

Furthermore, the integration of distributed cloud with intelligent computing networks presents a robust solution to
address the evolving challenges in autonomous driving technology. Through the dynamic allocation of computing
resources and the deep integration of cloud, network, and chip technologies, distributed cloud empowers
autonomous driving systems with enhanced data processing capabilities, ensuring stable and reliable performance
in various driving scenarios.

In conclusion, the synergy between distributed cloud, intelligent computing networks, and autonomous driving
technology marks a significant milestone in the advancement of intelligent transportation systems. As the demand
for autonomous driving technology continues to grow, the adoption of distributed cloud solutions is poised to
accelerate, driving innovation and transformation across the automotive industry.

ACKNOWLEDGEMENT
I would like to extend my sincere gratitude to the authors of two influential studies that have significantly inspired
and guided my research. First, my thanks go to Zhou, Yanlin, et al. for their insightful paper, [5]"Utilizing
AI-Enhanced Multi-Omics Integration for Predictive Modeling of Disease Susceptibility in Functional
Phenotypes," published in the Journal of Theory and Practice of Engineering Science. Their innovative approach
to AI-enhanced multi-omics integration has provided a valuable foundation for my work.
Additionally, I am deeply appreciative of the contributions made by Wang, B., He, Y., Shui, Z., Xin, Q., and Lei, H.
in their study on [6]"Predictive Optimization of DDoS Attack Mitigation in Distributed Systems using Machine
Learning." Their research has greatly influenced the development of my methodologies and has been a key source
of inspiration for my project.

Thank you for your groundbreaking research and the significant impact it has had on my work.

REFERENCES
[1] Xu, J., Wu, B., Huang, J., Gong, Y., Zhang, Y., & Liu, B. (2024). Practical Applications of Advanced Cloud
Services and Generative AI Systems in Medical Image Analysis. arXiv pr
[2] eprint arXiv:2403.17549.
[3] Zhang, Y., Liu, B., Gong, Y., Huang, J., Xu, J., & Wan, W. (2024). Application of Machine Learning
Optimization in Cloud Computing Resource Scheduling and Management. arXiv preprint arXiv:2402.17216.
[4] Gong, Y., Huang, J., Liu, B., Xu, J., Wu, B., & Zhang, Y. (2024). Dynamic Resource Allocation for Virtual
Machine Migration Optimization using Machine Learning. arXiv preprint arXiv:2403.13619.
[5] Zhou, Yanlin, et al. "Utilizing AI-Enhanced Multi-Omics Integration for Predictive Modeling of Disease
Susceptibility in Functional Phenotypes." Journal of Theory and Practice of Engineering Science 4.02 (2024):
45-51.
[6] Wang, B., He, Y., Shui, Z., Xin, Q., & Lei, H. Predictive Optimization of DDoS Attack Mitigation in
Distributed Systems using Machine Learning.
[7] Wu, Binbin, et al. "Enterprise Digital Intelligent Remote Control System Based on Industrial Internet of
Things." (2024).
[8] Pan, Y., Wu, B., Zheng, H., Zong, Y., & Wang, C. (2024, March). THE APPLICATION OF SOCIAL
MEDIA SENTIMENT ANALYSIS BASED ON NATURAL LANGUAGE PROCESSING TO CHARITY.
In The 11th International scientific and practical conference “Advanced technologies for the implementation
of educational initiatives”(March 19–22, 2024) Boston, USA. International Science Group. 2024. 254 p. (p.
216).
[9] Li, X., Zheng, H., Chen, J., Zong, Y., & Yu, L. (2024). User Interaction Interface Design and Innovation
Based on Artificial Intelligence Technology. Journal of Theory and Practice of Engineering Science, 4(03),
1-8.

21
 2024 World Journal of Innovation and Moder
ISSN 2682-5910
ernn Tec
echhnology, Vol.7, Issue 3 (Jun)

[10] He, Zheng, et al. "Application of K-means clustering based on artificial intelligence in gene statistics of
biological information engineering."
[11] Wu, Y., Jin, Z., Shi, C., Liang, P., & Zhan, T. (2024). Research on the Application of Deep Learning-based
BERT Model in Sentiment Analysis. arXiv preprint arXiv:2403.08217.
[12] Shi, C., Liang, P., Wu, Y., Zhan, T., & Jin, Z. (2024). Maximizing User Experience with LLMOps-Driven
Personalized Recommendation Systems. arXiv preprint arXiv:2404.00903.
[13] Li, L., Xu, K., Zhou, H., & Wang, Y. (2024). Independent Grouped Information Expert Model: A
Personalized Recommendation Algorithm Based on Deep Learning.
[14] Che, C., Lin, Q., Zhao, X., Huang, J., & Yu, L. (2023, September). Enhancing Multimodal Understanding
with CLIP-Based Image-to-Text Transformation. In Proceedings of the 2023 6th International Conference on
Big Data Technologies (pp. 414-418).
[15] Huang, Zengyi, et al. "Research on Generative Artificial Intelligence for Virtual Financial Robo-Advisor."
Academic Journal of Science and Technology 10.1 (2024): 74-80.
[16] Tian, J., Li, H., Qi, Y., Wang, X., & Feng, Y. Intelligent Medical Detection and Diagnosis Assisted by Deep
Learning.
[17] Zhou, Y., Zhan, T., Wu, Y., Song, B., & Shi, C. RNA Secondary Structure Prediction Using
Transformer-Based Deep Learning Models.
[18] Zhou, Tong, et al. "Image retrieve for dolphins and whales based on EfficientNet network." Sixth
International Conference on Computer Information Science and Application Technology (CISAT 2023). Vol.
12800. SPIE, 2023.
[19] Zhou, Hong, et al. "Application of Conversational Intelligent Reporting System Based on Artificial
Intelligence and Large Language Models." Journal of Theory and Practice of Engineering Science 4.03
(2024): 176-182.
[20] Xu, Kangming, et al. "Intelligent Classification and Personalized Recommendation of E-commerce Products
Based on Machine Learning." arXiv preprint arXiv:2403.19345(2024).
[21] Zhou, Tong. "Improved sales forecasting using trend and seasonality decomposition with lightgbm." 2023
6th International Conference on Artificial Intelligence and Big Data (ICAIBD). IEEE, 2023.
[22] Shen, Xinyu, et al. "Biology-based AI Predicts T-cell Receptor Antigen Binding Specificity." Academic
Journal of Science and Technology 10.1 (2024): 23-27.
[23] Zheng, Haotian, et al. "Medication Recommendation System Based on Natural Language Processing for
Patient Emotion Analysis." Academic Journal of Science and Technology 10.1 (2024): 62-68.
[24] Chen, B., Zhu, Y., Ye, S., & Zhang, R. (2018). Structure of the DNA-binding domain of human myelin-gene
regulatory factor reveals its potential protein-DNA recognition mode. Journal of structural biology, 203(2),
170-178.
[25] Srivastava, S., Huang, C., Fan, W., & Yao, Z. (2023). Instance Needs More Care: Rewriting Prompts for
Instances Yields Better Zero-Shot Performance. arXiv preprint arXiv:2310.02107.
[26] Chen, B., Liu, Z., Perry, K., & Jin, R. (2022). Structure of the glucosyltransferase domain of TcdA in complex
with RhoA provides insights into substrate recognition. Scientific reports, 12(1), 9028.
[27] Chen, Baohua, Kay Perry, and Rongsheng Jin. "Neutralizing epitopes on Clostridioides difficile toxin A
revealed by the structures of two camelid VHH antibodies." Frontiers in Immunology 13 (2022): 978858.
[28] Ni, Chunhe, et al. "Enhancing Cloud-Based Large Language Model Processing with Elasticsearch and
Transformer Models." arXiv preprint arXiv:2403.00807 (2024).
[29] Yu, D., Xie, Y., An, W., Li, Z., & Yao, Y. (2023, December). Joint Coordinate Regression and Association
For Multi-Person Pose Estimation, A Pure Neural Network Approach. In Proceedings of the 5th ACM
International Conference on Multimedia in Asia (pp. 1-8).
[30] Xu, J., Zhu, B., Jiang, W., Cheng, Q., & Zheng, H. (2024, April). AI-BASED RISK PREDICTION AND
MONITORING IN FINANCIAL FUTURES AND SECURITIES MARKETS. In The 13th International
scientific and practical conference “Information and innovative technologies in the development of
society”(April 02–05, 2024) Athens, Greece. International Science Group. 2024. 321 p. (p. 222).
[31] Cai, G., Qian, J., Song, T., Zhang, Q., & Liu, B. (2023). A deep learning-based algorithm for crop Disease
identification positioning using computer vision. International Journal of Computer Science and Information
Technology, 1(1), 85-92.
[32] Cai, G., Zhang, Q., Liu, B., Jin, Z., & Qian, J. (2024). Deep Learning-Based Recognition and Visualization of
Human Motion Behavior. Academic Journal of Science and Technology, 10(1), 50-55.

22