Take the Deep Learning — Lecture 1: Introduction to ML and DL, Set 05: 20 multiple-choice questions (30 minutes) with instant scoring and a full answer…
The first 5 questions of this test, with the correct answer and explanation.
Q1. A thermostat switches the heater on below 18 degrees C. How is this classified in the quiz?
- A. Rules written by hand, not machine learning at all ✓
- B. Shallow machine learning
- C. Deep learning
- D. Reinforcement learning
Answer: A. Rules written by hand, not machine learning at all
Explanation: The answer given is A: a fixed rule, with nothing learned from examples.
Q2. A bank flags a card payment as suspicious using 30 hand-picked features such as amount, time and country. Why is this shallow machine learning rather than deep learning?
- A. Because the dataset is too small for a deep network
- B. Because it is trained on labelled transactions, but the features were engineered by people ✓
- C. Because banks are not allowed to use neural networks
- D. Because it produces a yes/no answer rather than a probability
Answer: B. Because it is trained on labelled transactions, but the features were engineered by people
Explanation: It is B: trained on labelled transactions, so it is machine learning, but the features were engineered by people rather than learned by layers.
Q3. A spreadsheet formula converts marks into grades. Why is this not machine learning?
- A. Because spreadsheets cannot run algorithms
- B. Because the marks are not stored as a tensor
- C. Because deterministic rules are applied, with no data and no loss function ✓
- D. Because it has only one input column
Answer: C. Because deterministic rules are applied, with no data and no loss function
Explanation: The answer is A: deterministic rules, no data, no loss function.
Q4. A chess engine built only from hand-coded rules belongs to which category?
- A. Deep learning, because chess is a complex task
- B. Machine learning, because it improves with play
- C. None of the three, because chess is not an intellectual task
- D. Artificial intelligence, but not machine learning and not deep learning ✓
Answer: D. Artificial intelligence, but not machine learning and not deep learning
Explanation: AI includes approaches that do not learn at all; symbolic chess programs were all rules and no learning, so they are AI but not ML or DL.
Q5. What question marks the beginning of AI in the 1950s, according to the timeline?
- A. Can computers think? ✓
- B. Can computers calculate faster than humans?
- C. Can computers store large amounts of data?
- D. Can computers replace factory workers?
Answer: A. Can computers think?
Explanation: The 1950s entry on the timeline is "Can computers think?", when the effort to automate intellectual tasks normally done by humans begins.
Take the timed test (free, no sign-up) to see your score and the answer and explanation for every question.
Q6. When did machine learning begin to flourish, and what did it become?
- A. In the 1950s, becoming the first branch of AI
- B. In the 1980s, becoming a branch of symbolic AI
- C. In the 1990s, becoming the most popular and most successful subfield of AI
- D. In the 2010s, becoming a replacement for statistics
Q7. How does the lecture describe the relationship between machine learning and statistics?
- A. Machine learning is a branch of statistics with a different name
- B. Machine learning rejects statistical methods entirely
- C. Statistics has been replaced by machine learning
- D. It is rooted in statistics and borrows its mathematics, yet differs from it in several important ways
Q8. What does the lecture say about the amount of mathematical theory in machine learning?
- A. It is fully grounded in proofs before any method is used
- B. There is comparatively little mathematical theory, maybe too little
- C. The theory is complete but too difficult to teach
- D. Theory is irrelevant because results are always reproducible
Q9. What does scalability mean in the list of reasons deep learning is simpler to use?
- A. It can be trained on any size of dataset without changes
- B. It uses less memory as the model grows
- C. It parallelises naturally on GPUs and TPUs
- D. It can be split across many separate research teams
Q10. What does the lecture say a student needs in order to start with deep learning?
- A. A dedicated server with multiple GPUs
- B. Access to a university supercomputing cluster
- C. A TPU purchased from a cloud vendor
- D. A student laptop plus free cloud GPUs such as Colab and Kaggle
Q11. What does the lecture say about layer-wise pre-training?
- A. It came and went
- B. It remains the standard way to initialise deep networks
- C. It was never tried in practice
- D. It replaced backpropagation entirely
Q12. Why are industry data partnerships listed as a source of support for research?
- A. They provide free GPUs to universities
- B. Hospitals, banks and agri-businesses have the data but not the models
- C. They guarantee publication in applied journals
- D. They pay higher salaries than academic posts
Q13. Which Pakistani funding sources are named for this kind of research?
- A. PSDP and the Planning Commission
- B. State Bank research grants and SECP funds
- C. HEC NRPU and Local Challenge Fund grants, and Ignite National Technology Fund
- D. Pakistan Science Foundation and PCSIR only
Q14. What does "a working mental model" mean as a course outcome?
- A. Being able to derive backpropagation from first principles
- B. Being able to recite the architecture of every major network
- C. Being able to implement a framework such as Keras from scratch
- D. Being able to explain in plain words what a model learns, why it fails, and what data it needs
Q15. By which week does the course promise you will train your first network on real data?
- A. Week one
- B. Week three
- C. Week eight
- D. The final week
Q16. What recipe does the course give for "a path to a paper"?
- A. Public datasets, a baseline, and a well-run experiment, inside one semester
- B. A novel architecture invented from scratch
- C. A large grant and a research team
- D. A systematic literature review of 100 papers
Q17. What is the first assignment, and how is it marked?
- A. Write a literature review of five papers; marked on depth of analysis
- B. Build a network from scratch in NumPy; marked on correctness
- C. Train a model on any public dataset using a free Colab notebook and bring one screenshot of its accuracy; no marks for accuracy, full marks for doing it
- D. Collect a local dataset of 1,000 images; marked on dataset size
Q18. What does the course say about attendance and about cheating?
- A. Attendance is compulsory and cheating carries a warning
- B. Attendance is highly recommended, and there is absolutely no cheating
- C. Attendance is optional and cheating is handled by the department
- D. Attendance counts for 80% and cheating is not mentioned
Q19. What does the homework component of the course consist of?
- A. Only programming assignments
- B. Only problem sets
- C. Group projects and presentations
- D. Problem sets, paper reviews and programming assignments
Q20. What kind of problems did symbolic AI solve successfully, and what did it never manage?
- A. It solved well-defined logical problems, but never found the rules for perception and language
- B. It solved perception problems, but never managed logic
- C. It solved language translation, but never managed chess
- D. It solved nothing of practical value