AI0-001 AI Concepts and Techniques Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Add more convolutional layers

  Why it's wrong here

  Adding convolutional layers increases model capacity, which worsens the train-validation gap once validation loss has already begun rising — the network is overfitting, not underfitting. Extra layers would help if both losses were still high and plateaued, indicating the model lacked capacity to capture the underlying patterns.

• ✗

  Increase the learning rate

  Why it's wrong here

  Raising the learning rate increases step size, which worsens the divergence between training and validation loss rather than closing it. A higher rate is correct when training loss stalls too early from overly slow progress, not when validation loss is already rising.

• ✓

  Implement early stopping

  Why this is correct

  Early stopping halts training at the epoch where validation loss begins rising, restoring the weights from the best validation checkpoint. This directly counters the stem's overfitting pattern, improving generalisation by preventing further memorisation of training data.

• ✗

  Reduce the batch size

  Why it's wrong here

  Batch size governs gradient noise and step count, not the growing gap between training and validation loss; shrinking it changes optimisation dynamics without directly countering overfitting. Smaller batches suit memory-constrained training or regularisation through noise, not this divergence.