2024 Course Recap

What is NOT supervised learning?

• 1. Training a model with labeled data -- giving it the expected answers!

• Generating text ( i.e. Chat GPT) in response to user input given past text

• 1. Classifying cats and dogs, given pictures of each with labels.

• 1. Predicting the price of a house given historical data about cost vs. sq ft

If our model uses this dotted line to represent the relationship between values and time, what is our model doing?

• Gerrymandering

• Underfitting

• Overfitting

• Dropping out

What is linear regression? (2 CORRECT)

• The process of shifting weights and biases to minimize loss

• Moving to the left along a line to optimize the learning rate.

• Approximating a continuous (linear) output; finding the relationship between X and y

• "Finding the line of best fit"

What is loss?

• A number indicating how bad the model’s prediction was on a single example.

• A line that separates classes

• A line that shows the relationship between input features and output

• A function that transforms any number into a probability

Why is gradient descent important?

• 1. We want to make our predictions smaller, so descending helps

• We use it to adjust the weights / biases to make our model more accurate

• We use it to prevent the model from overfitting the data.

• 1. Gradient descent increases loss. More loss helps our model get close to 0

True or False: Hyperparameters are parameters that can't be learned by the model. The engineer chooses the hyperparameters! Examples include: learning rate, # of neurons, # of training iterations, etc.

• True

• False

1. What is Mean Squared Error? (2 are correct!)

• 1. The gradient of the function

• 1. The absolute difference between actual value and predicted value

• 1. A loss function

• 

1. Why do we use logistic regression instead of linear regression?

• 1. Logistic regression fits a linear equation to our data!

• 1. It is more logical

• Logistic regression is supervised learning, and linear regression is not.

• 1. Logistic regression classifies our inputs!

1. What is an activation function?

• 1. A function that decides whether a neuron is activated or not

• A function that activates a motor

• 1. A function that determines whether or not the feature the neuron detects should have an influence over the end prediction.

• A function that tells you how many neurons to use in a layer

How can we avoid overfitting? [3 CORRECT]

• Use MORE data

• 1. Use forms of regularization, like L1 / L2 regularization or dropout

• Use LESS data

• Use a MORE diverse dataset