Show HN: Seq2Seq ML Learns the Inverse of Manual Multiplication (Gelosia Method) (gitlab.com)

A new deep learning project has emerged that trains a sequence-to-sequence (Seq2Seq) neural network to perform the inverse of manual multiplication using the ancient Gelosia (Lattice) method. This initiative uniquely challenges the model to deduce intermediate multiplication results from only the final product, emphasizing its capability for inference based on limited data. The process is divided into three phases: data generation, model training using Keras/TensorFlow with a Mean Squared Error loss function, and subsequent testing through an interactive command-line tool. The Gelosia method simplifies multiplication into a grid-based format, breaking down complex calculations into manageable single-digit multiplications. The project specifically targets the multiplication of two 3-digit numbers, producing a 3x3 grid of partial products. The model uses a straightforward architecture without attention mechanisms, with an encoder-decoder setup based on LSTM networks. This experimentation not only showcases the adaptability of neural networks in learning mathematical processes but also highlights potential implications for enhancing automated reasoning and inference in AI/ML systems. The full project, complete with extensive datasets and model files, is accessible on GitLab, inviting further exploration and contributions from the AI community.