Qwissit Artificial Intelligence

ISTQB CT-AI

ML Metrics and Neural Networks

60 questions

ML functional performance metrics

  1. Which four situations does a confusion matrix distinguish in a classification? Easy
  2. What does a false positive (FP) mean for a spam filter if "spam" is the positive class? Easy
  3. How is accuracy calculated? Easy
  4. How is precision calculated? Easy
  5. How is recall calculated? Easy
  6. Which other name is used for recall? Easy
  7. How is the F1-score defined? Easy
  8. A classifier yields TP = 40, TN = 50, FP = 5 and FN = 5. What is the accuracy? Easy
  9. A model has predicted 40 cases as positive, of which 30 are actually positive. What is the precision? Easy
  10. A test dataset contains 50 actually positive cases, of which the model identifies 45 as positive. What is the recall? Easy
  11. Why does a classification model rarely predict everything correctly all the time? Easy

Neural networks

  1. Which three main types of layers does a deep neural network have? Easy
  2. What can a single-layer perceptron be used for? Easy
  3. What are the neurons in the hidden layers also called? Easy
  4. What is an epoch when training a neural network? Easy
  5. How are the weights and bias values of a neural network typically initialized at the start of training? Easy
  6. Why are special coverage measures needed for neural networks? Easy
  7. What does neuron coverage measure? Easy
  8. What is the situation regarding commercial tools for neural network coverage measures? Easy
  9. What is the task of the output layer of a neural network? Easy

ML functional performance metrics

  1. A classifier yields TP = 80, FP = 20, FN = 40 and TN = 860. What are precision and recall? Medium
  2. A model achieves a precision of 80% and a recall of 60%. What is the F1-score (rounded)? Medium
  3. A classifier yields TP = 90, FP = 30, FN = 10 and TN = 70. Which values are correct? Medium
  4. Which question does precision answer? Medium
  5. What does a low F1-score mean? Medium
  6. A model for early cancer detection should miss as few actual cases of the disease as possible. Which metric is the priority? Medium
  7. A spam filter should sort out as few important emails as spam by mistake as possible. Which metric (spam = positive) is the priority? Medium
  8. What range of values applies to the F1-score? Medium
  9. A model yields TP = 45, FP = 5, FN = 15 and TN = 35. Which values are correct? Medium
  10. Which statement about the presentation of the confusion matrix is correct? Medium
  11. A model yields TP = 60, FP = 40, FN = 20 and TN = 880. What is the accuracy? Medium

Neural networks

  1. How does a neuron (after the input layer) calculate its activation value? Medium
  2. What happens during training with the error (loss) between the generated output and the known correct result? Medium
  3. Which statements about the bias of a neuron are correct? Medium
  4. How does a deep neural network process the input data on the way to the output? Medium
  5. How are fully connected networks and perceptrons related? Medium
  6. How does k-multisection neuron coverage (kMNC) work? Medium
  7. What does neuron boundary coverage (NBC) measure? Medium
  8. What are coverage measures for neural networks mainly used for? Medium
  9. Why is structural coverage alone not sufficient to assess a neural network? Medium

ML functional performance metrics

  1. Of 1,000 transactions, 10 are fraudulent. A model classifies all transactions as "no fraud" (fraud = positive). Which statements are correct? Hard
  2. A model yields TP = 8, FP = 2, FN = 2 and TN = 988 for 1,000 cases. Which values are correct? Hard
  3. Model A achieves precision 90% and recall 50%, model B precision 70% and recall 70%. Which model has the higher F1-score? Hard
  4. A model classifies all 1,000 cases as positive; 50 are actually positive. Which values result? Hard
  5. A model achieves a recall of 90%. The test dataset contains 200 actually positive cases. How many of them did the model miss? Hard
  6. A model has predicted 50 cases as positive and achieves a precision of 80%. How many false alarms (false positives) does this contain? Hard
  7. A confusion matrix is presented with swapped axes: row "actually positive": 70 predicted positive, 30 predicted negative; row "actually negative": 10 predicted positive, 890 predicted negative. Which values are correct? Hard
  8. In fraud detection, both missed fraud cases and false alarms are expensive. Which metric is suitable as a single summary criterion? Hard
  9. A model yields TP = 50, FP = 50, FN = 0 and TN = 900. Which values are correct? Hard

Neural networks

  1. Which logical function can a single-layer perceptron learn? Hard
  2. In which order do the steps of training a neural network take place? Hard
  3. Two otherwise identical networks differ only in their activation function. What follows from this? Hard
  4. Which statements about deep neural networks are correct? Hard
  5. A network has 200 neurons. During testing, the output of 150 neurons exceeds the specified threshold. What is the neuron coverage? Hard
  6. A network has 10 neurons whose output ranges are each divided into k = 4 sections. During testing, 28 sections are activated. What is the kMNC? Hard
  7. In a network with 50 neurons, the output of 5 neurons during testing lies outside the range observed in training (above the maximum or below the minimum). Which measure with which value results? Hard
  8. A test suite achieves 100% neuron coverage, yet the model fails in the field on slightly rotated images. Which explanations and consequences are there? Hard
  9. The test evaluation shows that a group of neurons in a hidden layer is never activated. How should a tester react? Hard
  10. A team wants to check whether its tests also cover behavior beyond the neuron outputs observed in training. Which coverage measure fits best? Hard
  11. A test manager wants to use classic statement and branch coverage as the main measure for a neural network. What speaks against this? Hard

Certifications & exams

Which exam comes next?

You can already practise for 19 exams, and 21 are up for a vote. See all certifications with their official information, cast your vote or suggest your exam.

See all certifications