Sort Model Problems by Symptom
Identify common neural-network failure modes from training, validation, and production symptoms.
Sort each symptom into the most likely neural-network failure mode. Underfitting Overfitting Data drift Leakage Label noise Training and validation performance are both weak Training loss falls but validation loss rises after epoch 6 Live accuracy drops after a new pricing policy changes ticket mix A feature uses information recorded after the prediction date Two reviewers label the same refund case differently Holiday traffic patterns appear that were absent from training A larger model improves training score but hurts validation score A simple baseline performs about as poorly as the neural net
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in