#error-analysis

Question 1: Single choice

Your training images are scraped from the web, but the validation and test images come from the phone app in which the model will run. You hold out part of the training set as a train-dev set, which the model never trains on. Accuracies: training 98%, train-dev 97%, validation 85%. What is the most likely diagnosis?

The web images differ from the app images (a data mismatch)1
The model overfits the training set2
The model underfits because it is too simple3
The learning rate is too high, so training has not converged4

Timed test