Machine learning, models, and AI systems.
Test your knowledge with these top questions!
Why is ordinary accuracy often unavailable for unsupervised clusters?
Without known labels, analysts judge clusters by their structure and how well they support later tasks.
Principal component analysis keeps its first axis because that direction captures the most:
The first principal component follows the greatest spread in the data, preserving its strongest variation first.
Which training pattern separates a random forest from boosted trees?
Independent trees can train in parallel; boosted trees usually must wait for the model built in earlier rounds.
A game-playing AI updates its strategy from wins and penalties. Which learning approach fits?
A policy acts as the agent's strategy for choosing actions, while rewards guide how that strategy improves.
What does unsupervised learning usually lack in its training data?
With no answer key, the algorithm looks for structure such as clusters or compact representations.
Which values does a neural network mainly adjust while learning?
Weights scale each connection, while biases shift a neuron's response before the next layer.
A spam model returns 0.8. What turns that score into a yes-or-no label?
A threshold acts like a cutoff: scores above it enter the positive class, while lower scores enter the negative class.
In k-means clustering, what does the user normally choose before training?
The chosen k sets the group count; k-means then moves centroids and reassigns points until the solution settles.
Repeated model tweaks based on test scores effectively turn the test set into what?
Repeatedly consulting test scores leaks test-set knowledge into development and weakens the final evaluation.
In an image classifier, pixel values describe each photo. What role do those values play during training?
Features are the clues a model examines; text, sound, and tabular data supply their own kinds of clues.