Final Stuff

Question 1

MSE and MAE formulas

Answer 1

true - predicted, square/take absolute value, sum up, divide by n

Question 2

advantages of MSE and MAE

Answer 2

MSE - differentiable, good for learning
MAE - result is interpretable, simple, less prone to outliers

Question 3

accuracy

Answer 3

correct vs total
diagonal vs total

Question 4

recall/true positive rate

Answer 4

also called true positive rate
TP / (TP + FN)
out of all the positives, how many was the model able to get correct (recall)
first column

Question 5

false positive rate

Answer 5

FP / (FP + TN)
second column

Question 6

precision

Answer 6

TP / (TP + FP)
how many predicted positives were truly positive?
across first row

Question 7

how to set up confusion matrix

Answer 7

actual on top, predicted on the side
positive positive in the top left

Question 8

false negative

Answer 8

we predict negative, but that is false (actually positive)

Question 9

false positive

Answer 9

we predict positive, but that is false (actually negative)

Question 10

types of transformers

Answer 10

encoder, decoder, encoder-decoder

Question 11

high level of how to train an LLM

Answer 11

pretraining - predict next token
supervised fine tuning - train on prompts and good responses
reinforcement learning with human feedback - humans rate responses

Question 12

normalization vs regularization

Answer 12

normalization - making sure weights are same scale
regularization - makes sure model doesn’t overfit

Question 13

types of normalization

Answer 13

batch norm, L2 norm

Question 14

what are support vectors in SVM

Answer 14

the data points that are closest to the hyperplane

Question 15

what is out of bag evaluation

Answer 15

evaluate tree on the data that wasn’t used for training

Question 16

what is calibration?

Answer 16

making sure the output of the model represents how confident it is

Question 17

2 examples of non parametric models

Answer 17

knn, decision trees
don’t make assumptions about the underlying distribution

Question 18

what is LIME?

Answer 18

a technique that aims to create an interpretable model local to a data point

Question 19

what are proxy models?

Answer 19

models that behave similar to complex models

Question 20

why does the vanishing/exploding gradient problem occur?

Answer 20

we are multiplying gradients as we move to earlier layers, multiplying lots of small/large numbers together

Question 21

types of autoencoders

Answer 21

deep (multiple layers), sparse, variational, etc.

Question 22

what is the loss function for logistic regression?

Answer 22

binary cross entropy, it is strictly convex

Question 23

what are scaling laws?

Answer 23

take in model size and data and try to predict loss
how much will throwing resources at the model improve it?

Question 24

three svm kernels

Answer 24

linear, polynomial, RBF

Question 25

why use kernels in SVM?

Answer 25

transform the data into a higher dimensional space where the data is now linearly seperable

Question 26

what are sampling layers in cnns?

Answer 26

pooling layers

Question 27

what is grokking?

Answer 27

when test error falls way after training error has fallen, model suddenly generalizes well.