from-scratch-neural-network

A from-scratch implementation of a feedforward neural network in C# (.NET 8) without using any machine learning frameworks.

A convolutional neural network (CNN) built from scratch using only NumPy to classify handwritten digits from the MNIST dataset.

Deep feedforward neural networks in pure NumPy manual forward and backward propagation, object-oriented and functional APIs, weight-initialisation comparison, unit tested.

Part 3 of Visual Scrambling — four ResNet-style architectures reconstructed in PyTorch, with residual blocks, custom BatchNorm, theory notebooks, and CIFAR-100 experiments.

neural network from scratch. ive implemented backpropagation from scratch using micrograd, which i build by scratch using karpathy's guide. also has optimizer, and coded a little one some manual approached to optimize using randomizing weights and biases.

My first artificial neuron built from scratch in Python. It learns to approximate a linear function from data points using gradient descent, MSE, and manually calculated gradients, with a real-time visualization of the learning process.

A tiny deep-learning library built from scratch in pure NumPy; hand-written backprop verified by gradient checking. 99.6% on spirals, 97.8% on MNIST.

CNN MNIST / CIFAR10 Classifier - CNN Implemented From Scratch in CuPy/NumPy

Neural network from scratch to predict cirrhosis stages

Predicting clinical trial duration from registration data. Neural network built from scratch in NumPy, benchmarked against sklearn and Keras.

A GPT/Llama-style language model built from scratch in Rust — hand-written forward & backprop, every gradient numerically verified, trained on CPU, exported to GGUF for llama.cpp & Ollama.

A simple neural network built from scratch in Python to learn the XOR function. It implements forward propagation, backpropagation, sigmoid activation, gradient descent, and training without using machine learning libraries.

Train a handwritten digit recognizer from scratch using a custom-built Artificial Neural Network (ANN) in NumPy. This project generates the model weights (model_weights.npz) used by the DigitRecognizerWebApp repo. Includes full training pipeline, real image testing, and activation flow visualization.

A lightweight, end-to-end implementation of Stable Diffusion built from first principles on a single T4 GPU. Features a custom 192-channel U-Net, VAE, and a CLIP encoder, optimized for consumer hardware and trained on approx. 168k images.

ZeroProp is a custom-built, zero-dependency (excluding NumPy) neural network engine. It is designed to demonstrate a deep, foundational understanding of backpropagation, matrix calculus, and gradient descent without relying on abstractions from libraries like PyTorch or TensorFlow.

A minimal feedforward neural network

轻量 Transformer 文本分类器（路由模型）的训练尝试。复刻 Andrej Karpathy 的 nn-zero-to-hero 与 nanoGPT 架构，从零构建了一个 AI 文本分类器，旨在验证其在特定组件中进行意图识别的能力。