This repository documents a step-by-step attempt to solve the following problem:

Try to learn a neural network adding two binary numbers or performing multiplication by two (note that result should also be some binary number).

This project focuses on the first part of the task: learning binary addition with a neural network.

This is not only a final solution. It is a record of the whole process:

• understanding the task,
• building the program step by step,
• testing and improving the code,
• training a neural network in PyTorch,
• evaluating the results,
• asking what should come next.

• task/ – original task description
• conversation/ – step-by-step project story
• code/ – 8 Python files showing how the program evolved
• results/ – selected outputs and observations
• next_steps/ – open questions and future directions

For 4-bit binary numbers, the MLP model achieved perfect accuracy on all 256 possible input pairs.

Does this approach still work well for larger binary numbers, or do we need a different architecture that better captures binary carry propagation?