Binarized NN with MaxSAT

Overview

This project trains Binarized Neural Networks (BNNs) using MaxSAT, encoding the problem as a SAT instance for optimization.

Implementation

• BNN without hidden layer: Direct mapping.
• BNN with hidden layer: Improved non-linearity handling.

Results

Model	Solver Cost	Train Acc	Test Acc
BNN (No Hidden Layer)	27	0.75	0.54
BNN (Hidden Layer)	2	1.00	0.80