Neural Network in C++

A neural network with Qt GUI for training and visualization. Includes XOR gate and support for more datasets (e.g. handwritten digits).

Usage (GUI)

(Run ./NeuralNetworkGUI)

1. Topology – Configure layers (default: 2-4-1 for XOR)
2. Training Data – Browse to select a .txt file (e.g. data/xor.txt or data/digits.txt)
3. Load from file – Load topology from the training file
4. Create Network – Build network from topology
5. Train – Train on the selected data (Do not set the epchos to high on big data sets or many neurons since the training is running on your CPU it will likely freez the application)
6. Test / Predict – Enter inputs and run a forward pass
7. Click input neurons – Edit values directly in the visualization

With the digits dataset a small 8×8 drawing canvas is available for live prediction. Accuracy hovers around 90%; not ideal, but impressive for a from scratch implementation. Recognition is especially strong for digits like 8, even with distorted or incomplete inputs.

Requirements

• cmake and make
• Qt for GUI

Project Structure

Neural-Network-CPP/
├── src/
│   ├── core/           # Neural network (Net, Neuron, Connection, TrainingData)
│   └── gui/            # Qt UI (MainWindow, NetworkScene, NeuronItem)
├── tools/              # Training data generators
│   └── generateXorData.cpp
├── data/               # Training data files
│   └── xor.txt
├── main.cpp            # CLI entry point
├── main_gui.cpp        # GUI entry point
└── CMakeLists.txt

Build

mkdir -p build && cd build
cmake ..
cmake --build .

GUI only (Release):

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --target NeuralNetworkGUI

macOS (Qt):

cmake -B build -DCMAKE_PREFIX_PATH=$(brew --prefix qt@6)
cmake --build build

Run

GUI (recommended):

# macOS (.app bundle)
open build/NeuralNetworkGUI.app

# Linux / other (plain executable)
./build/NeuralNetworkGUI

CLI:

# Run from project root so data/xor.txt is found
./build/NeuralNetCLI

Generate Training Data

XOR gate:

./build/GenerateXorData > data/xor.txt

Handwritten digits (8×8, 10 classes):

./build/GenerateDigitsData > data/digits.txt

Uses embedded synthetic patterns (500 samples). For the full 1,797-sample UCI dataset, download optdigits.tes from UCI, place in data/, and run:

./build/GenerateDigitsData data/optdigits.tes > data/digits.txt

Data Format

topology: <input_count> <hidden_1> <hidden_2> ... <output_count>
in: <value1> <value2> ...
out: <target1> <target2> ...