This is the CoAI personal research repository (GitHub) — a collection of deep learning model implementations from scratch across multiple domains, organized for learning and experimentation. There is no build system, test framework, or CI/CD; this is not a library or application. Each top-level directory is a Python package (contains __init__.py) representing a research domain:
| Directory | Domain | Own-code highlights |
|---|---|---|
nlp/ |
Natural Language Processing | Transformer, LSTM, ResNet, ViT from scratch |
cv/ |
Computer Vision | DINO v1/v2, activation functions, normalization layers, MLP |
vlm/ |
Vision-Language Models | CLIP from scratch (full implementation in vlm/clip/) |
vla/ |
Vision-Language-Action | Autonomous driving VLA models (early stage) |
rl/ |
Reinforcement Learning | PPO implementation (rl/ppo/) with custom buffers/policies |
agc/ |
Adaptive Generative Control | DiT (Diffusion Transformer) models |
3d/ |
3D Vision | Placeholder |
agent/ |
AI Agents | Empty — planned for future work |
leetcode/ |
Algorithm Practice | Jupyter notebooks with classic algorithm problems |
Latest AI Papers: A quick overview of the latest papers on artificial intelligence, which you can check out Latest AI Papers to learn about cutting-edge artificial intelligence technologies.