Projects

Forum Database Using Spring Boot and Vue

Frontend: The Vue framework dynamically renders and updates HTML pages, ensuring responsive and interactive user interfaces. Its component-based development approach makes the code modular, maintainable, and improves development efficiency and readability.
Backend: Spring Boot is used to develop controllers, services, and data access layers for business logic and database operations. Its simple configuration and robust ecosystem streamline application development. RESTful APIs enable seamless real-time communication between the frontend and backend.
Database: MySQL 8.0.27 is employed as the relational database, offering stable and high-performance data storage. Using Spring Data JPA, database operations are simplified, making CRUD operations intuitive and efficient. MySQL’s advanced features, such as transaction management, index optimization, and stored procedures, further enhance system performance and data consistency. Reproduction of classic papers on Group Evolution Game Theory
Code

This project is a reproduction of the classic paper “Scale-Free Networks Provide a Unifying Framework for the Emergence of Cooperation” and “Social diversity promotes the emergence of cooperation in public goods games” by F. C. Santos. We reproduce the results of the paper and also explore the influence of different parameters on the model and the impact of different initial conditions on the final results.
The Code is updating…

Customized Spatial-Temporal data (Check-in Sequence) for LLMs enables these models to generate predictions or insights based on patterns and trends identified in the sequence of check-ins. By feeding LLMs with such data, it becomes possible to model complex human mobility and activity patterns over time and space, allowing for enhanced applications in areas such as targeted advertising, event prediction, and personalized recommendation systems.
Code

This model introduces a novel spatial-temporal lossless encoding method that effectively encodes spatial and temporal characteristics into a unified format. Furthermore, we utilize two disentangled diffusion modules that are conditioned on each other to capture the correlations between spatial and temporal aspects of check-in sequences, while reducing the impact of noisy interference. Finally, we incorporate contrastive learning to enhance the relationship between the two diffusion modules.

This project is based on the KITTI dataset and the competition private dataset, and the goal is to detect the 3D objects in the point cloud. The project is implemented by using the TED model. We apply sparse convolutional backbones to extract multi-channel transformed equivariant voxel features and the final result is evaluated by the mAP metric.
Code

This project is based on the Jeff dataset, and its goal is to predict the winner of the tennis match point by point and mining momentum changes. The project is implemented by using the Transformer model, and the final accuracy rate is as high as 94%.

This is a reproduction of the paper: UTNet: A Hybrid Transformer Architecture for Medical Image Segmentation. Integrated attention into neural networks for heart imaging segmentation. By applying self-attention modules in both the encoder and decoder, the model captures remote dependencies of different scales, enabling a more nuanced understanding of spatial relationships within cardiac images.
Code

This is a small C compiler coded in Python. Including lexical analysis, syntax analysis, semantic analysis, intermediate code generation, and target assembly code generation.
Code