Panik

Question 1

table-lookup agent

Answer 1

mapping indexed by percept sequences. Each action is just an entry in that table.

Question 2

reflexive agent

Answer 2

condition-action rules that are pattern matched against a percept.

Question 3

Goal-oriented agents

Answer 3

try to fulfil (binary) goals by imagining the outcome of actions.

Question 4

Utility-based agents

Answer 4

have a utility function that judges the expected outcomes of different actions somehow.

Question 5

Learning agents

Answer 5

improves the system by changing and retrieving knowledge from the performance element

Question 6

accessible

Answer 6

All relevant aspects of the world are available to the sensors.

Question 7

deterministic

Answer 7

The next state depends completely on the current state and chosen action.

Question 8

episodic

Answer 8

The choice of an action depends only on the current state (not on the past).

Question 9

static

Answer 9

The world does not change while deciding on an action.

Question 10

discrete

Answer 10

There are only finitely many world states in a range.

Question 11

successor function S ( x )

Answer 11

returns the set of states reachable by any action from state x.

Question 12

path

Answer 12

sequence of actions

Question 13

Single-state problem
Multiple-state problem
Contingency problem
Exploration problem

Answer 13

World knowledge | Action knowledge
complete | complete
incomplete | complete
to be found at run-time | incomplete
unknown | unknown

Question 14

Completeness:

Answer 14

Always finds a solution if it exists.

Question 15

(blind) search

Answer 15

uninformed

Question 16

(heuristic) search

Answer 16

informed

Question 17

Iterative deepening

Answer 17

combines BFS and DFS. It executes depth-limited search at increasing depths until
a solution is found.

Question 18

Non-complete search?

Answer 18

Iterative Deepening

Question 19

The heuristic h used for A* must be admissible, that is

Answer 19

h ( n ) ≤ h* ( n ) must hold where h* ( n ) is the cost of an optimal path from n to the nearest goal.

Question 20

To compare the quality of admissible heuristics

Answer 20

we introduce the effective branching factor b*

Question 21

Iterative Deepening A*

Answer 21

a variant of Iterative Deepening search that explores branches up to a given threshold for the value of f ( n ). If this threshold is passed and no solution was found the threshold is set to the minimal value of f ( n ) for all found nodes n that exceeded the threshold.

Question 22

Simplified Memory-Bounded A*

Answer 22

uses a bounded priority queue. New nodes are added to the queue. Each time a node is added, the algorithm checks whether all sibling nodes are in the queue, then
the parent is removed. If the memory is full, the node with the highest cost is removed from the queue and its parent is added (if not already present).

Question 23

Hill Climbing

Answer 23

iteratively expands the highest-valued successor of the current node.

Question 24

Games are

Answer 24

special cases of search problems. States are usually accessible. Actions are possible moves by a player.

Question 25

Knowledge Level:

Answer 25

What is known by the knowledge base.

Question 26

Symbolic Level:

Answer 26

The encoding of the knowledge base in a formal language.

Question 27

Implementation Level:

Answer 27

Internal representation of sentences, like lists or strings of things.

Question 28

Deductive Inference

Answer 28

a process to compute the logical consequences of a KB, i.e. given a KB and a sentence alpha we want to compute if KB |= alpha.

Question 29

correct

Answer 29

the computation is correct, e.g. if we compute KB |= a then KB |= a.

Question 30

complete

Answer 30

for every sentence a such that KB |= a we can also compute that KB |= a.

Question 31

Skolemisation is not equivalence preserving but

Answer 31

satisfiability preserving.

Question 32

The Most General Unifier (MGU) can be calculated using the following algorithm in

Answer 32

exponential time.

Question 33

Planning

Answer 33

Given a set of actions, an initial and a goal state, find a plan to reach the initial state from the goal state. Such a plan will consist of an arrangement of (possibly only partially) ordered actions.

Question 34

Planning VS Search

Answer 34

uses more detailed information e.g. preconditions and not just edges

Question 35

complete plan

Answer 35

requires that for any step every precondition to be fulfilled by some predecessor and that no step between the fulfilment and the requirement of a condition undoes this.

Question 36

consistent plan

Answer 36

requires that no two actions take place at the same time.

Question 37

solution

Answer 37

a plan that is both complete and consistent

Question 38

Partially ordered plans

Answer 38

preorder on steps which means it defines a must-happen-before-relationship

Question 39

d-Separation time

Answer 39

polynomial

Question 40

d-Separation complete?

Answer 40

no

Question 41

supervised learning

Answer 41

the learner has access to input and correct output.

Question 42

Reinforcement learning

Answer 42

only gives feedback in terms of rewards and punishment, but not correct answers.

Question 43

Unsupervised learning

Answer 43

happens without any feedback to the learner. The learner must learn on its own.

Question 44

Feed-forward networks

Answer 44

directed acyclic graph

Question 45

Recurrent networks

Answer 45

arbitrarily complex connections