Whenever you're ready, walk me through how you'd approach this.
How would you improve the performance of a neural network?
Assume you are given a working neural-network baseline, training data, validation data, and an evaluation script. Explain and demonstrate a systematic process for diagnosing underfitting, overfitting, optimization problems, data issues, and unsuitable architecture choices. Your solution should include reproducible Python code and justify how you would measure whether each change improves the model rather than merely fitting the validation set.