This repository contains implementations of encoder-decoder recurrent neural networks (RNNs) using Long Short-Term Memory (LSTM) cells to perform basic arithmetic operations (addition, subtraction, multiplication) on two-digit integers. The project explores sequence-to-sequence learning, employing three different tasks including text-to-text, image-to-text, and text-to-image models on the MNIST dataset for image representations. For each task, we investigated the generalization capability of the implemented models in performing arithmetic operations. The analysis focuses on model accuracy, character-level misclassifications, and the nature of the mistakes made by the model.
For detailed methodology, results, and discussion, refer to the included PDF reports.