AI/ML Basics
Name something that uses/is both AI and ML.
Many possible examples. Most common types/uses of AI also use ML.
What is Learning Rate?
A value that affects how quickly a model learns in training
In y=mx+b, what is m?
The slope
What is the lowest (deepest) node of a tree called?
Leaf node
Explain what the Turing Test is.
Whether an AI's output can be determined from a human, originally defined to be in the context of a conversation
What is Loss?
A value describing how wrong a model's prediction is.
What function does logistic regression try to fit?
Sigmoid function
Identify one use of a decision tree.
Many possible answers
Does a self-driving car use AI, ML, or both? Explain.
Both - ML is involved in training AI models that detect objects and drive the car.
No, they are not the same - accuracy measures the raw percentage of correct responses. Loss is function that determines how wrong outputs are (lower is better).
Magnus and Hikaru are playing a chess match. To predict the outcome, do you use linear or logistic regression? Explain.
Logistic regression - since the outcome can be represented as a probability between 0 and 1
In a decision tree, what do the leaf nodes represent?
Outcomes
In Pac-Man, do the ghost enemies use AI, ML, or both? Explain.
AI only. The ghosts are not trained to learn, so they do not use ML.
During training, does a higher loss affect how much a model's parameters move? If so, do the model's parameters move more or less?
Higher loss increases change in parameters.
To predict stock prices, do you use linear or logistic regression? Explain.
Linear regression - continuous values
How many roots does a decision tree have? Is it always constant?
Does a robot necessarily use AI or ML? (e.g. the ones you might see in robotics competitions) Explain.
No. If the robot is entirely driven by humans, it does not use AI or ML.
Which scenario would you prefer for training a model? Using a dataset with 100,000 entries but 20% are incorrectly labeled, or using a dataset with 50,000 entries but all are correctly labeled?
Explain why.
The 50k entries dataset - data quality is VERY important for model training, as models can only be as good as the data they are trained on.
Is it more likely for linear regression or polynomial regression to overfit to data? Explain. (Polynomial regression is a more complex type of regression.)
Think carefully!
Polynomial regression. Generally, more complex models are more likely to overfit, due to being better at fitting data tightly.
Explain whether setting a very high decision tree depth is a good idea.
Bad idea - leads to overfitting