👨‍💻 I'm an AI enthusiast currently exploring the fascinating world of deep learning and its applications. My interests lie in Computer Vision (CV) and Large Language Models (LLMs).

🌱 I’m currently sharpening my skills in PyTorch and diving deeper into model optimization and deployment.

💡 I enjoy transforming ideas into real-world projects and contributing to open source. Feel free to explore my repositories below!

📫 How to reach me: 18350282676@163.com or xmkevin2004@gmail.com

Programming Languages:

ML/DL Frameworks:

Tools & Platforms:

1. Flask 2048 Enhanced - Tech: Python, Flask, Jinja2, HTML/CSS/JS

   A Flask-based 2048 web game with classic move/merge logic, featuring undo (up to 20 steps), poison tiles (clear after consecutive stays), countdown tiles (halve value each step), score/high score tracking, step/time statistics, and achievement system (64-8192).

2. Love in Stars - Tech: Three.js, WebGL, Canvas

   A romantic starry particle animation webpage with dynamic starry background, meteor effects, and particle text/graphics transformations. Features four particle modes, background music controls, and dual-canvas layering technology.

3. macOS Editor Demo - Tech: Electron, Monaco Editor, JavaScript, HTML/CSS

   A minimal runnable editor with macOS native interactions: trackpad gestures (two-finger tab switching/pinch-to-zoom), macOS-style context menus, Dock badges, Touch Bar buttons, and system notifications.

1. Adversarial Lunar Lander - Tech: PyTorch, Gymnasium, Box2D, Stable-Baselines3

   LunarLander environment with random/adversarial wind/gravity disturbances, upgraded to POMDP. Uses Recurrent PPO + LSTM for robust landing strategies in harsh dynamic environments.

2. CLIP Reproduction - Tech: PyTorch, Transformers, ResNet50, BERT, Contrastive Learning

   Lightweight CLIP model reproduction with dual-tower architecture (ResNet-50 image encoder + BERT text encoder), trained with contrastive learning (InfoNCE/CLIP Loss) for multimodal representation, supporting zero-shot classification and image-text retrieval.

3. Hybrid Recommendation Engine - Tech: Python, NumPy, Pandas, Scikit-learn, Apache Spark MLlib

   Hybrid recommendation system integrating user/item-based collaborative filtering, K-Means/DBSCAN clustering for user segmentation, and matrix factorization (SVD) for latent feature discovery and improved prediction accuracy.

1. MCM A - Drone Smoke Bomb Optimization - Tech: NumPy, Pandas, Matplotlib, Concurrent Processing

   Mathematical Contest in Modeling A: Physical simulation + evolutionary search to optimize multi-drone smoke bomb timing/parameters for maximum missile concealment duration. Uses UNION metric with parallel multi-restart, grid comparison, and visualization.

2. CPU Simulator - Tech: Python

   Feature-rich 8-bit CPU simulator with complete instruction set, interrupt handling, memory-mapped I/O, and floating-point operations.

3. Easy Compiler - Tech: Python, C++

   Simple compiler that translates custom language or C-like subset to assembly code.

4. 24C++ Endterm Project - Tech: C++

   Self-developed music rhythm game (work in progress).

Machine Learning:

• Perceptron - Perceptron model implementation
• SVM - Support Vector Machine implementation
• KNN - K-Nearest Neighbors algorithm
• Machine Learning EndTerm - Final course project

Deep Learning:

• Deep Learning 2nd - CNN experiments on Fashion-MNIST with LeNet, VGG, ResNet architectures
• Deep Learning 3rd - Titanic survival prediction using MLP with data preprocessing and feature engineering
• Deep Learning 4th - Advanced deep learning experiments

Algorithms & Data Structures:

• Kruskal-Prime - Kruskal and Prim's algorithms for minimum spanning trees
• Titanic Predict - Kaggle competition: Predict survival on the Titanic

1. Cutting-edge Intelligent Algorithms - Course materials and final paper
2. Operations Research 1 - Course slides and assignments
3. Operations Research 2 - Assignment solutions and notes