MNIST Digit Recognition with Neural Networks in C++ This project implements a neural network from scratch in C++ to recognize handwritten digits from the MNIST dataset. The Eigen library is used for matrix operations, enabling efficient computation in training and prediction phases.

Overview The project trains a neural network to classify images of handwritten digits (0-9) from the MNIST dataset. The neural network is fully implemented in C++, with the Eigen library handling matrix operations, which allows the model to be trained without relying on deep learning frameworks.

Features Neural Network: Multi-layer perceptron (MLP) with configurable layers and neurons. Eigen Library: Used for efficient matrix operations. MNIST Compatibility: Trained on the MNIST dataset for digit recognition. From Scratch Implementation: No deep learning libraries, only Eigen for linear algebra.

Load Dataset Set the File Paths in Code: Update the file paths in main.cpp to point to the extracted files. For example: std::string train_image_file = "C:\Users\username\Desktop\MNIST\train-images.idx3-ubyte"; std::string train_label_file = "C:\Users\username\Desktop\MNIST\train-labels.idx1-ubyte"; std::string test_image_file = "C:\Users\username\Desktop\MNIST\t10k-images.idx3-ubyte"; std::string test_label_file = "C:\Users\username\Desktop\MNIST\t10k-labels.idx1-ubyte";

Results After training 10 epochs, the model achieves an accuracy of approximately 90,2% on the MNIST test set.