Things i should know

Question 1

What is the curse of dimensionality wrt supervised learning

Answer 1

The number of training examples needed to correctly learn increases exponentially as we increase the dimensionality of our inputs.

Question 2

Advantages of hierarchical clustering over K means and visa versa

Answer 2

• don’t need to know clusters in advance
• easy to understand
• Faster
• Easier to implement

Question 3

What are semantically different hypotheses?

Answer 3

All permutations of hypotheses being where each element can be x+1 things. x being its possible values. +1 being ?. We also add 1 to represent the specific hypothesis. i.e (x+1)(y+1)…+1

Question 4

What are syntactically different hypotheses?

Answer 4

All permutations of hypotheses being where each element can be x+2 things. x being possible instance values. +2 being ? and ø.

Question 5

Is a rote learner biased?

Answer 5

No. It is trained on the set of arbitrary disjunct hypotheses and is therefore not biased. This does mean that it has overfit the training data and will not be able to generalise

Question 6

What fundamentally classifies a markov decision process?

Answer 6

Past actions do not affect the transition function. The markov property is when the transition function is preserved throughout the entire process.