avatar

Xiangxiang Cui

A PhD student at BNU focuses on designing intelligent computing models.

About me

Hoping to build safe AI world models by studying the brain, complex systems, mathematics and physics, contributing to scientific progress and benefiting humanity.

InterestsEducation
⚡️Efficient AI‬
🏞Generative AI
🙇‍♂️Trustworthy AI
🧠Brain-inspired AI
🌍Multimodal LLM
🎓Ph.D. in Beijing Normal University, 2022-Present
State Key Laboratory of Cognitive Neuroscience and Learning
🎓M.E. in Xi’an JiaoTong University, 2019-2022
Faculty of Electronic and Information Engineering
I am dedicated to designing intelligent computing models/systems, focusing on developing meta-methods for reasoning, learning, security, and efficiency in AI systems. My interests span World Models, Multimodal AI, Physics AI, Continual Learning, Trustworthy AI, Generative AI, Large Language Models, Reinforcement Learning, Robotics, Multi-Agent Systems, NLP, Neuro-symbolic AI, Neural Decoding, Neural Coding, and AI Systems. If you are interested in any of these topics, please feel free to contact me. I am open to cooperation.

Activities

  • 🙇‍♂️Working on my Ph.D thesis.
  • 🧪Coding on Depression Analysis based on large-scale dataset.
  • 🧪Coding about Multi-modal classification of psychiatric disorders.

News

  • 🎉 2025/01 A patent (No. 202410125896.X) for the application of artificial intelligence in radiotherapy.
  • 🎉 2024/08 In collaboration with the Department of Oncology and Radiology of Peking University Third Hospital and other hospitals, we submitted a journal paper to the Medical physics 2024 (JCR Q1, Top 3 Medical Physics journals), Accepted.
  • 🎉 2024/04 A paper about eliminating site effects will be submitted to EMBC 2024, CAAI B, Accepted.
  • 🎉 2023/12 An abstract of neuroimage classification was submitted to OHBM2024, Poster presentation.
  • 🎉 2023/10 Join in the National key research and development program.

Selected Publications

​ # Equal contributions

  • Xiangxiang Cui#, Xueying Yang#, et al. “A StarGAN and Transformer-Based Hybrid Classification-Regression Model for Multi-institution VMAT Patient-Specific Quality Assurance.” In Medical Physics 2024 (JCR Q1, CAS Q2, Top 3 Medical Physics journals). [GAN|Site effect]
  • Xiangxiang Cui, Dongmei Zhi, Weizheng Yan, et al. “CGDM-GAN: An Adversarial Network Approach with Self-supervised Learning for Site Effect Removal. “ In EMBC 2024, CAAI B. [GAN|Self-supervised learning|Site effect]
  • Xiangxiang Cui, et al. “Multi-Scale Analysis Framework Integrating Brain Functional Connectivity and Activity” In the Annual Meeting of the organization for Human Brain Mapping (OHBM) 2024. [Classification|NeuroAI]
  • Xiangxiang Cui, et al. “CGDM-GAN: An Adversarial Network Approach with Self-Supervised for Removing Site Effects” In the Annual Meeting of the organization for Human Brain Mapping (OHBM) 2023. [Classification|Site effect]
  • Xiangxiang Cui, et al. “Variable-frame CNNLSTM for Breast Nodule Classification using Ultrasound Videos” Preprint 2023. [Videos classification]
  • Yichen Wang, Zhongyu Li, Xiangxiang Cui, et al. “Key-frame Guided Network for Thyroid Nodule Recognition using Ultrasound Videos” In MICCAI 2022. [Videos classification]
  • Xiangxiang Cui, et al. “DEAttack: A differential evolution based attack method for the robustness evaluation of medical image segmentation.” In Neurocomputing 2021, IF:5.779. [Model robustness|Segmentation|Adversarial attack]

Research experiences

Ultrasound Video Classification
  • Feature extraction of multiple consecutive video frames based on CNN and LSTM.
  • The verification accuracy of video-based(Our) classification is higher than single-frame classification.
This research was done in the Frontline Intelligent Technology Internship.
Head and Neck Segmentation
  • Label Data Preparation: using regional growth and optical flow to generate pre-segmentation. The Label specialists manually revised pre-segmentation.
  • Deep Learning: using label data and 3D CNN to generate smooth segmentation results.
This research was done in the ShenZhen Keya Medical Intership.
Coronary Segmentation
  • Using CNN to segment the aorta, analyse the aorta segmentation results and then adjust algorithmic to adapt to different hospital datasets.
  • Participating in the refinement of coronary segmentation, including vessel broken, vein removal.
  • Develop vessel segmentation results evaluation tools to improve iterative efficiency.
This research was done in the ShenZhen Keya Medical Intership.

Engineering experiences

Slurm Cluster Management(Centos)
  • Slurm configuration fix.
  • Multi-user data privacy protection.
  • Software environment configuration.
As a server administrator in the tutor lab of Beijing Normal University and CAS.
Multi-User Server Management(Ubuntu)
  • Deep learning, Docker environment configuration, system installation and problem fixes.
  • Multi-user data privacy protection. Each user has separate account access. Users are also grouped by project, and folders cannot be accessed between different groups.
  • Configure Remote Desktop Program so that every user can remotely access the remote desktop. Configure intranet penetration programs to ensure that the intranet can be accessed by external network programs under special circumstances.
As a server administrator in the tutor lab of Xi’an Jiaotong University.
Dance Robot
  • The robot consists of rudder, and motion instructions are issued by the microcomputer system through a serial port.
  • For the coordination and smoothness of the dance, we carefully design the dance moves and use the data structure to coordinate the control action set.
Web Online Compiler
  • The whole system is completed by using Html, Javascript, Php, Linux shell, Mysql.
  • Support languages: C, C++, C#, Java, Html, Javascript, Php.

The Embedded Car
  • The whole system is completed by using microsystem STM32, Cortex-A8 and C language.
  • Using KNN to identify direction(turn to left or right) based on the Opencv library.
  • Using PID algorithm and sensor to control the travel stability of the embedded car.

Union Search
  • Based on JS, HTML to implement multiple search engine(such as: google/bing/baidu.) split-screen results display.
  • This program is simple, but very practical, once online for a while.

Algorithm Greedy Snakes Game
  • Design the game interface and develop intelligent snake algorithms using C++/CLR Library.
  • Based on the traditional greedy snake algorithms, our algorithm takes food as the target and takes artificially controlled snakes as an obstacle to make path planning. In the end, the winning rate of the algorithm snake exceeds 70%.

Compus Today Android APP
  • Realize the functions of BBS forum, news and information download, etc.
  • This project participated in the Innovation and Entrepreneurship Competition for Chinese College Students.

Electronic Labeling Systems
  • The whole system is completed by using microsystem STM32, Wireless hardware module, Ajax, C, HTML, LWIP Network Protocol, ESP8266 module, HTTP Server.
  • Electronic labeling systems can be used in warehouses. The flow is sent from the web, via the wireless network module to the digital display, when the warehouse personnel completes the corresponding order, press the button of the electronic label, and finally the web will show that the task is complete.


Services

Conference Reviewers: ICLR2025