Unit 4: Evolution and Modern AI Paradigms - Practice Quiz

1 Which approach represents knowledge using rules and symbols?

2 Which approach learns patterns from data and probabilities?

3 What is the main purpose of machine learning?

4 What is a model in machine learning?

5 What type of data is used in supervised learning?

6 Which task is an example of supervised learning?

7 What is the main goal of unsupervised learning?

8 Which technique commonly groups similar data points?

9 How does an agent learn in reinforcement learning?

10 What does a reinforcement learning agent choose?

11 What is deep learning primarily based on?

12 Why are deep learning models called deep?

13 Which deep learning model is commonly associated with image recognition?

14 Which application commonly uses deep learning for understanding spoken words?

15 What is a foundation model?

16 What is one common way to adapt a foundation model?

17 What does multimodal Artificial Intelligence process?

18 What is a key feature of agentic Artificial Intelligence?

19 Where does Edge Artificial Intelligence process data?

20 What is a main characteristic of Tiny Artificial Intelligence?

21 A hospital must encode a policy stating that a patient is eligible for a treatment only if several explicit clinical rules are satisfied. Which AI approach is most suitable when the decision must be directly traceable to those rules?

22 A spam filter performs well during testing but becomes less accurate after spammers begin using new vocabulary. Which issue most directly explains this decline?

23 A company has historical houses with known selling prices and wants to predict the selling price of a new house. What type of supervised task is this?

24 A disease occurs in only 1% of patients. A classifier predicts every patient as healthy and obtains 99% accuracy. Which metric would better reveal its failure to identify diseased patients?

25 A retailer wants to group customers by purchasing behavior but has no predefined customer categories. Which method best matches this objective?

26 A dataset contains 100 strongly correlated sensor measurements per machine. An engineer wants a smaller set of features that preserves as much variation as possible. Which technique is most appropriate?

27 A warehouse robot receives for delivering a package, for a collision, and for each movement. Why is the small movement penalty useful?

28 An agent always chooses the action with the highest current estimated value and never tries alternatives. What problem is most likely?

29 Why can adding nonlinear activation functions between neural network layers increase the range of functions the network can represent?

30 A system must identify defects in images even when the defect appears at different positions. Which neural architecture is especially suitable?

31 A translation system must relate each generated word to relevant words anywhere in a long source sentence. Which mechanism best supports this requirement?

32 An organization adapts a pretrained language model to classify legal documents using a relatively small labeled legal dataset. Which process is being applied?

33 Why can one foundation model support summarization, question answering, and text classification?

34 A model answers questions about a chart by combining its title, labels, and plotted shapes. Why is this a multimodal task?

35 In a vision-language search system, what is the main benefit of mapping images and captions into a shared embedding space?

36 An AI assistant must schedule a meeting by checking calendars, proposing times, sending invitations, and retrying after conflicts. Which capability most clearly makes it agentic?

37 A purchasing agent can place orders autonomously. Which control most directly limits harm from an incorrect high-cost decision?

38 A factory camera must stop machinery within milliseconds when it detects a worker in a danger zone, even if the internet connection fails. Where should inference primarily run?

39 A neural network must run on a microcontroller with limited memory and power. Which change most directly reduces model storage by representing weights with fewer bits?

40 A battery-powered sensor must detect a wake word continuously using very little energy. Which deployment design best fits Tiny AI?

41 A diagnostic system must justify every conclusion using explicit domain rules, but its observations are noisy and incomplete. Which architecture best preserves symbolic interpretability while handling uncertainty?

42 A symbolic planner performs perfectly on known objects but fails when a new object is introduced, whereas a statistical model generalizes to the object but violates a hard safety constraint. What is the most appropriate synthesis?

43 A model achieves high accuracy on randomly split records, but its performance drops sharply when evaluated on patients admitted in a later year. Which diagnosis is most likely?

44 A model has excellent discrimination but produces probabilities that are systematically too high. Which evaluation and remedy are most directly appropriate?

45 In a medical classification task, positive cases are rare and false negatives are substantially more costly than false positives. Which evaluation strategy is most defensible?

46 A feature is recorded after a loan application is approved and is strongly predictive of repayment. Including it during training produces impressive validation results. What is the central problem?

47 K-means produces different clusters under feature rescaling, while the underlying relationships are unchanged. Which explanation is correct?

48 A clustering algorithm finds three groups with strong internal separation, but the groups disappear after removing a variable that encodes hospital identity. What should be concluded first?

49 An agent receives a reward only when it reaches a distant goal, and learning is nearly absent because exploratory trajectories rarely reach it. Which intervention most directly addresses this issue?

50 In Q-learning, an agent repeatedly selects an action with an optimistic estimated value, even after receiving poor outcomes. Which issue can cause this behavior when updates are infrequent?

51 Why can residual connections improve the optimization of a very deep neural network?

52 A neural network performs well on training data but poorly on unseen data, and increasing its depth worsens the gap. Which intervention is most directly aimed at the observed problem?

53 A convolutional model detects a small manufacturing defect, but the defect's location is needed for robotic repair. Which output task is required?

54 A speech system must convert an audio stream of variable duration into text of variable length while using surrounding context. Which model family is most naturally suited?

55 A foundation model performs well on a downstream task with a few labeled examples but fails when prompts contain a domain-specific abbreviation absent from pretraining. Which limitation is most relevant?

56 A language model generates a fluent citation that does not exist. Which system-level mitigation is most appropriate for a high-stakes retrieval question?

57 A multimodal model answers a question about a chart using text that contradicts the plotted values. Which capability is most likely deficient?

58 An agent can call external tools and is asked to delete duplicate records. Which design most reduces the risk of irreversible unintended actions?

59 An edge vision system must respond within 30 milliseconds even when network connectivity is intermittent. Which architecture best satisfies the latency requirement?

60 A model for a battery-powered sensor must fit within a strict memory budget and process streaming data continuously. Which optimization is most appropriate?