formulas

Question 1

what is enough data for PAC learning?

Answer 1

m >= (1/eps)*(ln(1/delta) + ln|H|)
m = num of samples
eps = average classification error
delta = probability of existence of a wrong hypothesis consistent with all examples. The risk that you are willing to take that just fits the data for the wrong reasons

Question 2

hoefdings formula:

Answer 2

m >= -ln((delta/2) / (2E^2))

Question 3

what is the vapnik chervonenkis dimension?

Answer 3

the max number of classes that a learning algo can learn to separate (shatter)

Question 4

what does bellman equation do?

Answer 4

relate the value function to itself via the problem dynamics

Question 5

formula of Q-value

Answer 5

Q(s,a) = r(s,a) + gammaV(delta(s,a))

delta: state * action -> state’

Question 6

what is the expected value of hte duration d in state i?

Answer 6

1/(1-a_ii)