NameGen

Character-Level Name Generator with Manual Backpropagation

A character-level name generator implemented in PyTorch, with a manual implementation of backpropagation.

The model learns patterns in names and generates new names by predicting one character at a time based on a fixed-length context.

• Paper: "Neural Probabilistic Language Model" by Bengio et al. 2003.
• It uses data from the U.S. Social Security Administration's Baby Names dataset
• Features a manually implemented backpropagation algorithm

Neural Architecture:

• Embedding Layer for character encoding.
• Flatten Layer
• Hidden Layer with Batch Normalization and Tanh activation.
• Output Layer for character prediction.

Dataset:

Processes over 100,000 unique names from the SSA Baby Names dataset.

Name Generation:

Produces unique, realistic names character by character.

Output Examples

priah
kailas
sabirg
kyena
deka
kailas
sabirg