Difficulty: Easy & Basic
Within AI, what does a rational agent seek to maximize?
A Memory footprint
B Code length
C Sensor count
D Expected utility
Answer: D. Expected utility. A rational agent selects actions to maximize expected utility in its environment.
Machine learning is commonly treated as a subfield of what?
A Artificial intelligence
B Compiler theory
C Database design
D Network routing
Answer: A. Artificial intelligence. Machine learning is a major approach within artificial intelligence.
Supervised learning uses which kind of training examples?
A Labeled examples
B Random actions
C Untagged clusters
D Hand-coded rules
Answer: A. Labeled examples. Supervised algorithms learn from examples paired with target labels.
When should a classifier predict?
A Random rewards
B Continuous measurements
C Discrete class labels
D Nearest coordinates
Answer: C. Discrete class labels. Classification predicts categories such as spam or not spam.
Regression usually estimates what?
A Class names
B Cluster IDs
C Continuous numeric values
D Action sequences
Answer: C. Continuous numeric values. Regression models predict numeric quantities such as temperature or price.
Unsupervised learning typically receives data without what?
A Numerical values
B Input features
C Data records
D Target labels
Answer: D. Target labels. Unsupervised methods find structure without supplied target labels.
Clustering groups which examples?
A Rule statements
B Labeled answers
C Sorted file names
D Similar data points
Answer: D. Similar data points. Clustering groups observations according to similarity in their features.
Model parameters are fitted using which data split?
A Class list
B Live stream
C Training set
D Test set
Answer: C. Training set. The training set is used to fit model parameters.
Final unseen-data evaluation relies on which split?
A Feature list
B Training set
C Test set
D Output layer
Answer: C. Test set. A held-out test set estimates performance on unseen examples.
Hyperparameter selection mainly uses which data split?
A Generating rewards
B Selecting hyperparameters
C Storing raw files
D Fitting final labels
Answer: B. Selecting hyperparameters. Validation data helps compare configurations during model development.
When does a model overfit?
A It lacks input features
B It memorizes training noise
C It has no parameters
D It predicts every class
Answer: B. It memorizes training noise. Overfitting occurs when a model fits training details that do not generalize.
Feature is the name for what kind of model input?
A Reward
B Label
C Epoch
D Feature
Answer: D. Feature. A feature is an input variable used by a model.
The target output in supervised learning is called what?
A Kernel
B Label
C Gradient
D Feature
Answer: B. Label. A label is the target value associated with a training example.
Layers of connected computational units form which architecture?
A Hash table
B Routing tree
C Neural network
D File system
Answer: C. Neural network. Neural networks connect computational units in layers.
Nonlinearity can be introduced by what network component?
A Nonlinearity
B Class balance
C More labels
D Training examples
Answer: A. Nonlinearity. Nonlinear activations let stacked layers represent nonlinear functions.
Once training ends, inference means what?
A Generating predictions
B Choosing a dataset
C Writing source code
D Labeling training data
Answer: A. Generating predictions. Inference applies a trained model to inputs to produce outputs.
Processing human language belongs to which field?
A Natural language processing
B Database normalization
C Signal routing
D Image compression
Answer: A. Natural language processing. Natural language processing studies computational tasks involving human language.
Image and video analysis belongs to what field?
A Rule chaining
B Data indexing
C Computer vision
D Speech synthesis
Answer: C. Computer vision. Computer vision extracts or interprets information from visual data.
Inside an expert system, where are domain facts and rules stored?
A Search frontier
B Training split
C Loss function
D Knowledge base
Answer: D. Knowledge base. An expert system’s knowledge base contains facts and rules.
Rules are applied to known facts by which component?
A Inference engine
B Output label
C Feature scaler
D Input sensor
Answer: A. Inference engine. An inference engine derives conclusions by applying rules to facts.
Search heuristics estimate what quantity?
A Number of labels
B Cost to goal
C Training duration
D Model accuracy
Answer: B. Cost to goal. A search heuristic estimates remaining cost or distance to a goal.
Breadth-first search expands which frontier node first?
A Random node
B Shallowest node
C Newest goal
D Highest-cost node
Answer: B. Shallowest node. Breadth-first search expands the shallowest nodes before deeper ones.
Depth-first search follows which frontier path first?
A Deepest available path
B Lowest estimated cost
C Widest layer
D Nearest goal only
Answer: A. Deepest available path. Depth-first search pursues a path deeply before backtracking.
Reward feedback guides which learning agents?
A Feature vector
B Reward signal
C Validation score
D Class label
Answer: B. Reward signal. Rewards provide feedback about the consequences of an agent’s actions.
Turing’s imitation game asks which question?
A Processor speed
B Sensor accuracy
C Database size
D Whether replies seem human
Answer: D. Whether replies seem human. The classic Turing test asks whether a machine’s responses can be distinguished from a human’s.