How would you diagnose overfitting with only 120 independent examples?
Instruction: A deep model achieves 100% training accuracy and 70% accuracy on 20 validation examples. A regularized linear baseline gets 75% on the same validation examples. Explain what this comparison supports and how you would choose the next experiment.
Context: Reason about model capacity and noisy validation without inventing statistical certainty.
Updated
Official answer available
Read the opening below, then unlock the full answer and practical guidance.
The deep model has a large train-validation gap, but 70% versus 75% is 14 versus 15 correct predictions. One additional correct example is insufficient by itself to establish that the linear model is reliably better...