In this exercise, you have a set of multiple choice questions. In each question, only one of the given options is correct, and only one can be selected.

1. A reactive agent:

a) Integrates sensory information over time

b) Reacts to environment stimuli in real time

c) Generates plans using state transitions described in the STRIPS format

d) Represents its knowledge using propositional logic

e) None of the previous

2. PDDL is a language for:

a) Representation of actions for planning

b) Representation of semantic networks with default knowledge

c) Representation of general purpose knowledge in first order logic

d) Exchange of knowledge between agentes

e) None of the previous

3. Graph-search differs from tree-search in that:

a) The graph-search uses semantic networks to represent the states

b) In the tree generated by graph-search, each state appears at most once

c) Graph-search is complete and optimal

d) In graph-search, transitions are always given in the STRIPS format

e) None of the previous

4. In natural language processing, the verb subcategorization mechanism allows to:

a) Represent gender agreement constraints between subject and verb

b) Represent number agreement constraints between subject and verb

c) Represent the syntactic arguments of a verb

d) Determine the semantic representation of a sentence

e) None of the previous

5. Hill-climbing search is:

a) A tree search technique for optimization problems

b) A random search technique in graph

c) An iterative improvement search technique that ends at the global maximum

d) A research technique whose probability of accepting negative improvements decreases over time

e) None of the previous

6. A robot is being developed for underwater fishing, Nautilus. Your collaboration is requested in the development of the control module. This robot carries a maximum of 10 harpoons and a deposit for 20 fish. When Nautilus feels a fish in front, it immediately throws a harpoon (action Throw). If the harpoon hits a fish, the robot stores the fish in the deposit (Store), in this case recovering the harpoon. Otherwise, the harpoon is lost. When all the harpoons are lost, Nautilus can get a new set of 10 harpoons (action GetHarpoons). When the fish deposit is full, the fish must be unloaded to an inland station (action Unload), leaving the deposit empty again. When Nautilus has nothing else to do, it just wanders around (action Wander).

a) Identify and characterize a set of predicates that can be used to describe situations in which Nautilus can be found. Identify also the state variables, if any are needed.

b) Specify a set of situation-action rules to govern the behavior of Nautilus. You can do this in the form of a table with the following columns:

- Situation - a conjunction of conditions in first-order logic;

- Update - update of state variables, if there are any;

- Action - action to be taken by the agent in the specified

7. Consider a domain made up of animals, species and instants of time, in which knowledge can be described through the following predicates:

- Animal(a): a is an animal

- Species(a,s): animal a belongs to species s

- Alive(a,t): animal a is alive in time t

- Extinct(s,t): species s is extinct in time t

- Parent(p,a): animal p parent of animal a

a) Represent the following sentences in first-order logic:

- Any animal has a parente

- Any animal belongs to a species

- Only animals belong to species

- If p is the parent of a and a belongs to a species s then p also belongs to s

- A species s is extinct in the time t if no animal of this species is alive in this time

- There were no live mammoths in the year 1918

b) Consider the following grammar for natural language processing:

sentence --> np, vp.

np --> artigo, noun.

article --> [the].

noun --> [sparrow].

noun --> [sparrows].

noun --> [dove].

noun --> [doves].

vp --> verb.

verb --> [flies].

verb --> [fly].

This grammar recognizes as correct the sentences [the, dove, fly] and [the, sparrows, flies]. Present an improved version of this grammar in which this problem is avoided.

The following grammar is augmented to produce the semantics of sentences:

sentence([in(X,Y)|SEM]) --> np(X^WHAT), [is, in], np(Y^WHERE),

{ append(WHAT,WHERE,SEM) }.

np(X^[belongs(X,Y)|SEM]) --> np_simple(X^SEM1), [of], np_simple(Y^SEM2),

{ SEM = [SEM1,SEM2] }.

np(X^[SEM]) --> np_simple(X^SEM).

np_simple(SEM) --> [the], noun(SEM).

noun(X^member(X,car)) --> [car].

noun(X^member(X,professor)) --> [professor].

noun(X^member(X,workshop)) --> [workshop].

The professor is in the workshop

The car of the professor is in the workshop

10 Consider the following grid world. The u values specify the utilities after convergence of the Value Iteration algorithm. The r value is the reward associated with the center state. Assume that ?= 1 and that an agent can perform four possible actions:

North, South, East and West. With probability 0.7, the agent reaches the intended state, with probability 0.2 it moves to the right of the intended direction, and with probability 0.1 to the left. Which is the best action na agent can execute if he is currently in the center state of the grid world?

Justify your answer. Which utility does the center state have?

u=8; u=15; u=12

u=2; r=2; u=10

u=7; u=16; u=11

Describe briefly the main experimental techniques for evaluating the performance of supervised learning algorithms for classification tasks.