fengbingchun/NN_Test

What it solves

This repository provides a comprehensive collection of deep learning and machine learning implementations in C++ and Python, serving as a practical reference for building neural networks and implementing mathematical foundations from scratch.

How it works

The project is split into two primary environments:

• C++ Implementations: Focuses on low-level implementations of mathematical formulas (linear algebra, probability, activation and loss functions), optimization algorithms (SGD, Adam, etc.), and classic ML models like Perceptrons, CNNs, and Decision Trees. It also includes integration with OpenCV DNN, libtorch, and ONNX Runtime for YOLOv8/YOLO11.
• Python Implementations: Provides high-level demos and scripts for modern AI frameworks. This includes RAG (Retrieval-Augmented Generation) using Ollama and LangChain, object detection and segmentation using Ultralytics (YOLO), and OCR using PaddleOCR.

Who it’s for

• Developers wanting to see how ML algorithms are implemented at a C++ level.
• AI Engineers looking for quick-start scripts for RAG, YOLOv8/11, and model quantization/pruning.
• Students studying the mathematical foundations of deep learning.

Highlights

• Dual-Language Support: Implementations in both C++ (for foundations) and Python (for modern frameworks).
• Broad Algorithmic Coverage: Covers everything from basic Linear Regression to advanced YOLOv11 instance segmentation.
• MLOps Tools: Includes scripts for model pruning, PTQ quantization, and knowledge distillation.
• RAG Integration: Ready-to-use examples for Ollama and LangChain RAG pipelines.