• Achieved precise segmentation of log bark regions by leveraging SAM for semantic segmentation, using automatically labeled positive (wood core) and negative (bark) sample points

• Proposed and implemented a log explosion diagram based on force-repulsion iteration to achieve 2D diffusion projection of 3D log point clouds, effectively preventing overlap between logs

• Improved cluster refinement performance by replacing CPU-based KD-Tree with FAISS GPU-accelerated nearest

neighbor search, resulting in significantly faster processing

• Built end-to-end intrusion detection models on NSL-KDD dataset using PyTorch and Scikit-learn

• Designed Bi-LSTM and Transformer Encoder architectures for sequential anomaly detection

• Implemented label encoding, normalization, and sliding-window sequence generation for temporal modeling

• Achieved 73.9% test accuracy, outperforming Transformer baseline by 24% in detection precision

• Visualized attention weights and confusion matrices to interpret model behavior and failure patterns

• Proposed SMOTE and focal-loss techniques to handle class imbalance across 22 attack categories

• Processed U.S. Census datasets (NC & FL) to clean and standardize millions of name-ethnicity records

• Optimized ML models (XGBoost, LightGBM, CNN, AdaBoost), achieving 75.67% accuracy after tuning

• Improved CNN architecture (dropout, filter sizes, optimizers), boosting accuracy by ~10%

• Visualized classification performance using confusion matrices, loss curves, and feature importance plots

• Established a Convolution Neural Network based on 30,000 tumor pictures using PyTorch, achieved accurate identification

of tumor edges and determined malignancy based on their contours

• Optimized the Deep Q Network (DQN) to identify the optimal coefficient for the binary polynomial

• Developed a full-stack school database system using JavaScript for both front-end and back-end

• Built the front-end with HTML, CSS, and vanilla JS, providing role-based portals for teachers and students

• Created the back-end with Node.js and Express.js, allowing secure grade upload and retrieval

• Integrated a MongoDB database to store user credentials, course records, and grades with proper access control

• Implemented role-based authentication via session management for secure data access

• Applied best practices in modular design, documentation, and testing to improve maintainability