Build a basic feed-forward network in PyTorch

Define a simple PyTorch nn.Module with one hidden layer. Write the training loop including the forward pass, Binary Cross Entropy (BCE) loss calculation, and an Adam optimizer step. Train for 50 epochs on quality_binary and report final test metrics.

No file or test is named in the issue; start by locating where quality_binary is defined and where the model experiments live. Use the specified PyTorch training requirements as the scope, and consider the work complete when 50 epochs have run and final test metrics are reported.

Written by the indexing model from the issue text.