Week 2

Question 1

Explain function approximation

Answer 1

Finding a pattern from examples so we can make good guesses for new data. Like learning that bigger houses tend to cost more, we create a rule to predict prices of houses we haven’t seen yet.

Question 2

What is a “feature” in a decision tree?

Answer 2

Something we know about the options or examples.

When we want to choose a restaurant features are type of food, price category etc.

Question 3

What are weight in AI neural networks and how are these weights determined

Answer 3

Weights work as classifiers and are determined by supervised learning so it learns to recognize patterns

Question 4

Explain back propagation in artificial neural networks

Answer 4

Process to improve accuracy: if the network’s guess is wrong, we adjust the weights backward, layer by layer, to reduce future mistakes.

Question 5

Why are AI neural networks effective?

Answer 5

1) They can understand complex patterns by organizing information in a hierarchy
2) They can handle huge amounts of data
3) They can sometimes outperform experts (diagnosing skin conditions)

Question 6

Explain universal approximation theorem

Answer 6

With the right structure and enough neurons, a neural network can learn almost any continuous function (pattern).

However, finding the right setup (topology) and training it correctly is key to making it work well.

Question 7

How do you make predictions via K-NN for classification and regression

Answer 7

1) Classification: most common label (= plurality vote) > 3 for dog, 4 for cat > it is a cat
2) Regression: Mean of the closest neighbor (= arithmic mean) > 200, 250 en 300 > (200+250+300)/3 = 250

Question 8

What are three goals of unsupervised learning?

Answer 8

1) Understand underlying structure
2) Reducing he dimensionality of a data set (describe, summarize, simplify) using less data
3) Organise similar data in groups or clusters