Cram2

Question 1

Why is Planning Different from Search?

Answer 1

• Search generates all successor states.

- Search wants to find a sequence of actions leading to a goal. Planning looks for a description of a plan.

Question 2

A Plan consists of

Answer 2

• a set of partially ordererd < plan steps
• a set of variable assignments x = t
• a set of causal relations

Question 3

Complete Plan

Answer 3

Every precondition of every plan step is satisfied

Question 4

Consistent Plan

Answer 4

If Si < Sj, then Sj -< Si and if x = A, then x != B

Question 5

P(A or B)

Answer 5

P(A) + P(B) - P(A and B)

Question 6

P(A | B)

Answer 6

P(B)

Question 7

P(A and B)

Answer 7

P(A | B) * P(B)

Question 8

Bayes: P(B | A)

Answer 8

P(A)

Question 9

P(Y | X,E)

Answer 9

P(X | E)

Question 10

d-separation

Answer 10

• can be computed in polynomial time
• is incomplete, that is, not every conditional independence is detected
• is nevertheless sufficient for a number of inference algorithms

Question 11

Goal of Learning

Answer 11

Optimize future behavior on the basis of the history of percepts, actions, and knowledge about the world.

Question 12

Supervised Learning

Answer 12

Both the input and the correct output are available to the learner

Question 13

Reinforcement Learning

Answer 13

While the correct answer is not available, there is feedback in terms of rewards and punishment

Question 14

Unsupervised Learning

Answer 14

There is no indication of what the correct output is

Question 15

Every propositional formula (Boolean function) is representable by

Answer 15

a decision tree

Question 16

Evaluating a Learning Algorithm

Answer 16

Keeping the training and test sets separate is crucial!

Question 17

Strategy of the Current Best Hypothesis

Answer 17

Only consider one hypothesis at a time. If there is a new example which is inconsistent with the hypothesis, then change it in the following way:

• generalization: make the extension bigger for a false negative example
• specialization: make the extension smaller for a false positive

Question 18

Some problems with the current-best hypothesis:

Answer 18

• All previous examples need to be tested again.

- uncontrolled backtracking

Question 19

Decision Lists

Answer 19

Decision lists (DL’s) consist of a number of tests, which themselves consist of a conjunction of a bounded number of literals. If a test is successful (all the 
literals are satisfied), then the DL tells us which value to return. Otherwise, the next test is tried.
A bit worse than DT.

Question 20

Advantages of Neural Networks

Answer 20

• Parallelism -> speed
• robustness
• graceful degradation
• inductive learning

Question 21

Feed-forward topology

Answer 21

DAG

Question 22

Recurrent topology

Answer 22

arbitrary directed graphs

Question 23

With 1 hidden layer

Answer 23

every continuous function is representable

Question 24

With 2 hidden layers

Answer 24

every function is representable

Question 25

Net is too big:

Answer 25

Overfitting

Question 26

XOR is

Answer 26

not representable when using a step function

Question 27

What are NNs good for?

Answer 27

attribute-based representations, in particular also for attributes with continuous values

Question 28

Minuses of NNs

Answer 28

no transparency

Bad at using additional knowledge

Question 29

P (occ_x,y)

Answer 29

Initialize the occupancy values of the grid to 0.5, that is, initially the robot has maximum uncertainty about the occupancy values.

Question 30

Markov Assumption

Answer 30

refers to conditional independence, e.g. a sensor reading only depends on the current location and the occupancy of the cells

Question 31

Weak AI

Answer 31

machines which act as if they were intelligent

Question 32

Strong AI

Answer 32

machines with a real (human like) conscious mind