Neural Networks
Name an application of a neural network
Many e.g. computer vision
Name a regression model.
Many possible answers e.g. linear regression
Define an ensemble model.
Model that combines predictions from other types of models
Name something that uses ML.
Many possible answers e.g. LLMs
What is the input of a neural network called?
Input layer
Logistic regression can be used to predict probabilities. True or false?
True
What do leaf nodes represent in a decision tree?
Class predictions
What is loss?
Value describing how wrong a model's prediction is
A deep neural network has a lot of this type of layer.
Hidden layer
What loss function is typically used in linear regression?
Mean squared error
Is random forest bagging, boosting, or stacking?
Bagging
Is image detection supervised or unsupervised learning?
Supervised
You have a neural network to recognize digits. What might hidden layers represent?
Specific features (eg loops, curls) of numbers
Does linear regression tend to overfit or underfit? Explain
Underfit; only works for very simple data due to its low complexity
In gradient boosting, each tree predicts ...
The loss of previous predictions
What is feature engineering?
Making new features (things the model uses) from data.
What would happen if a neural network didn't have a nonlinear activation function?
It would essentially reduce to a linear regression model.
Can logistic regression be used to predict stock prices? Explain.
No - predicts values between 0 and 1.
Which overfits more easily: random forest or a single decision tree? Explain why.
Single decision tree - random forest averages predictions
You are given a training dataset with 50,000 rows and 10 columns. Should you use all columns when training a model? Explain
No - some may contain things like ID, name, etc. that do not influence prediction, and cause overfitting