Deep Learning Assessment: Sample Questions | Bryq
Beginner
Intermediate
Advanced
These are three sample questions from Bryq’s Deep Learning assessment, a technical test for hiring deep learning engineers, machine learning engineers and AI researchers. The full assessment covers three levels: beginner, intermediate and advanced, from neural network architecture and components to training mechanics, model evaluation and practical applications. Each question is multiple choice, with one correct answer out of four options.
Beginner
Question
A training set has 10,000 samples, and the model is trained with a batch size of 50. How many iterations (weight updates) make up one epoch?
Select your answer
200
50
500
10,000
WHY IS THIS THE CORRECT ANSWER
One epoch is a full pass through the training data, and each iteration processes one batch, so the number of iterations is 10,000 ÷ 50 = 200. Batch size controls how many samples are used for each weight update, not how many passes are made over the data.
Intermediate
Question
While training a model, training loss keeps falling, but validation loss starts to rise after epoch 10. What should you do?
Select your answer
Stop training and keep the model weights from around epoch 10.
Keep training, since validation loss will eventually fall again.
Increase the learning rate to speed up convergence.
Remove the validation set, since it’s slowing training down.
WHY IS THIS THE CORRECT ANSWER
Validation loss rising while training loss keeps falling is a sign of overfitting: the model is memorizing the training data rather than learning patterns that generalize. Early stopping keeps the weights from the point where validation performance was best. Regularization techniques such as dropout, or more training data, can also help.
Advanced
Question
A convolutional layer receives a 64×64 input and applies a 3×3 kernel with a stride of 2 and padding of 1. What is the spatial size of the output?
Select your answer
32×32
64×64
62×62
31×31
WHY IS THIS THE CORRECT ANSWER
Output size = ⌊(input + 2 × padding − kernel) ÷ stride⌋ + 1 = ⌊(64 + 2 − 3) ÷ 2⌋ + 1 = 31 + 1 = 32. The stride of 2 roughly halves the spatial size, while the padding of 1 offsets the shrinkage that the 3×3 kernel would otherwise cause.
