QR Decomposition for Handwriting Recognition

This repository examines handwriting recognition task on EMNIST Letters dataset using QR decomposition .

Project Overview

• Dataset Preparation: For each class, the first 200 samples are used as training data, forming a data matrix for that class. An additional 20 samples per class are used as a test set to evaluate the model's performance.

• Part 1 – QR Decomposition with Householder Matrices:

  • Construct the data matrix for each class using the training samples.
  • Compute the QR decomposition of the matrix using Householder transformations.
  • Solve the least squares problem to predict labels for the test data and evaluate model performance.

• Part 2 – Incremental QR Decomposition with Givens Rotations:

  • Simulate adding 20 new samples per class incrementally, equivalent to appending columns to the data matrix.
  • Instead of recomputing the QR decomposition from scratch, update it using Givens rotations.
  • Solve the updated least squares problem to predict test labels and compare results with the previous method.

This project demonstrates how QR decomposition can be applied both in batch and incremental learning scenarios for handwriting recognition task.