Take the Deep Learning — Lecture 1: Introduction to ML and DL, Set 01: 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. AlphaFold, crop-disease diagnosis from a photo, Urdu speech-to-text, fraud detection and ChatGPT are given as five examples. What do they have in common?
- A. They all run on the same hardware
- B. Nobody wrote the rules; each system learned them from examples ✓
- C. They were all released in the same year
- D. They all require an internet connection
Answer: B. Nobody wrote the rules; each system learned them from examples
Explanation: The slide's answer is that nobody wrote the rules: each system learned them from examples.
Q2. How is artificial intelligence defined in the lecture?
- A. Any system that uses a neural network
- B. Any system that learns from data
- C. Any effort to automate intellectual tasks, including hand-written rules that never learn ✓
- D. Any computer program that can pass a human exam
Answer: C. Any effort to automate intellectual tasks, including hand-written rules that never learn
Explanation: AI is the general field: any effort to automate intellectual tasks, including hand-written rules that never learn.
Q3. Machine learning is defined as systems that are:
- A. Trained on examples rather than programmed with rules ✓
- B. Programmed with rules rather than trained on examples
- C. Built only from many successive layers
- D. Designed by experts without any data
Answer: A. Trained on examples rather than programmed with rules
Explanation: Machine learning covers systems trained on examples rather than programmed with explicit rules.
Q4. Deep learning is defined in the lecture as:
- A. Machine learning that uses very large datasets
- B. Machine learning that runs on a GPU
- C. Any AI system that beats a human
- D. Machine learning with many successive layers of learned representations ✓
Answer: D. Machine learning with many successive layers of learned representations
Explanation: Deep learning is machine learning with many successive layers of learned representations.
Q5. What is the correct nesting of the three fields?
- A. AI sits inside machine learning, which sits inside deep learning
- B. Deep learning sits inside machine learning, which sits inside AI ✓
- C. Machine learning sits inside deep learning, which sits inside AI
- D. All three are separate fields with no overlap
Answer: B. Deep learning sits inside machine learning, which sits inside AI
Explanation: The diagram shows deep learning inside machine learning, which sits inside artificial intelligence.
Take the timed test (free, no sign-up) to see your score and the answer and explanation for every question.
Q6. During the symbolic AI era of the 1950s to 1980s, how were systems built?
- A. Experts hand-crafted large sets of explicit rules
- B. Large neural networks were trained on labelled data
- C. Systems learned statistical structure from examples
- D. Programs were evolved by genetic algorithms
Q7. Which group of problems is described as the wall that symbolic AI hit in the late 1980s?
- A. Arithmetic, sorting and searching
- B. Chess, draughts and logic puzzles
- C. Image classification, speech recognition and translation
- D. Database queries and spreadsheet calculations
Q8. In the 1990s, what new approach replaced symbolic AI?
- A. Larger teams of rule-writing experts
- B. Faster hand-coded chess engines
- C. Quantum computing
- D. Letting the computer learn the rules from data
Q9. The lecture says a machine-learning system is trained rather than explicitly programmed. What happens during that training?
- A. It is shown many relevant examples and finds the statistical structure in them
- B. It is given a complete rulebook written by an expert
- C. It randomly guesses until it produces the right answer once
- D. It copies the behaviour of an existing symbolic AI program
Q10. In the holiday-photos example used to illustrate training, what does the system learn?
- A. How to take better photographs
- B. Which pictures go with which tags
- C. How to compress images more efficiently
- D. How to sort photos by date
Q11. The lecture calls machine learning hands-on and empirical. What does that mean in practice?
- A. Every result must first be proved as a theorem
- B. Only mathematicians can contribute to the field
- C. Ideas are proven by running experiments far more often than by proofs
- D. Results are decided by expert opinion rather than measurement
Q12. Why does the lecture say classical Bayesian analysis becomes impractical for modern machine-learning problems?
- A. Bayesian methods have been mathematically disproved
- B. Bayesian methods cannot handle labelled data
- C. Bayesian software is no longer maintained
- D. The scale is too large: millions of images, each with tens of thousands of pixels
Q13. What three things does the lecture say learning needs?
- A. Input data points, examples of the expected output, and a way to measure how well it does
- B. A GPU, a dataset, and a programming language
- C. A hypothesis, an experiment, and a publication
- D. Weights, layers, and an optimizer
Q14. What serves as the feedback signal used to adjust the algorithm?
- A. The size of the training dataset
- B. The number of layers in the network
- C. The gap between the current output and the expected output
- D. The time taken to process one example
Q15. According to the lecture, which step is what we actually call learning?
- A. Collecting the input data
- B. Labelling the expected outputs
- C. Choosing the measure of success
- D. The adjustment step
Q16. In the toy problem with black and white points, what are the inputs, expected outputs and measure of success?
- A. The colours; the coordinates; the training time
- B. The coordinates; the colours; the percentage of points classified correctly
- C. The number of points; their colours; the number of layers used
- D. The image pixels; the point labels; the loss value
Q17. After the coordinates of the toy problem are changed, what single rule solves it?
- A. Black points have y > x; white points have y < x
- B. Black points are closer to the origin than white points
- C. Black points have x > 0; white points have x < 0
- D. Black points outnumber white points
Q18. What is the hypothesis space?
- A. The memory available to store the training data
- B. The set of all datasets a model has seen
- C. The range of possible accuracy scores
- D. A predefined set of operations such as coordinate changes, linear projections and non-linear operations
Q19. Why does the lecture say machine-learning algorithms are not creative?
- A. They only search through the operations we allow in the hypothesis space
- B. They can only copy answers from their training data
- C. They cannot be applied to new problems
- D. They always produce the same output for any input
Q20. Machine-learning models are described as being all about finding:
- A. The fastest algorithm for a given hardware platform
- B. Useful representations of the input data that make the task easier
- C. The largest possible number of training examples
- D. A set of rules an expert can read and verify