The ultimate goal of the Robocup tournament is
- To build a robotic team that will beat the World Cup (humans) champions.
- To create intelligent agents.
- To deliver autonomous agent to Mars.
- To create a set of rules which will govern the behaviors of robots so that they will behave in acceptable ways.
- To give graduate students something productive to do with their time.
The Soccerserver communicates with the players using
In the soccerserver, if a player sends more than one
move command in one time slice, what happens?
- Only one of the commands will get executed.
- The server averages out all the commands.
- The player loses its turn.
- The soccerserver crashes.
- The soccerserver sends an error message back to the player.
In the soccerserver, the
sense_body contains what?
- Information about the player's stamina, neck, view_mode, etc.
- Commands for the soccerserver to change the player's stamina, speed, etc.
- Commands for the soccerserver to change the player's mode from sprint to marathon and viz.
- Information about the player's location.
- There is no such message.
Which one of the following is a good way to slow down the
performance of your RoboCup player?
- Place a lot of
println() in your code.
- Use hash tables.
- Replacing doubles with ints whenever possible.
- Implementing a subsumption architecture.
The ISIS RoboCup team learned that "competition with
collaboration" is a good technique. What did they mean by
- Allow players to compete with each other but only on certain occasions.
- That developers in one team should cooperate with developers on other teams.
- The best form of competition is collaboration.
- Nothing. Its just a phrase that sounds deep but has no meaning.
- That by collaborating with each other the players on the team could better compete against the other teams.
Which one of the following items only has common value?
- a share of stock
- a house
- an ice cream cone
- a car
- a signed edition of "The Neuromancer" by William Gibson
What is the dominant strategy for an English auction?
- Bid slightly above the current price until you reach your reservation price.
- Wait until everyone seems to stop bidding and then bid.
- Do not participate because under all circumstances the English auction leads to a winner's curse.
- Bid up to your reservation price plus some epsilon.
- Bid half of your reservation price, then double whatever anyone else bids.
One problem with a First-Price Sealed-Bid auction is that
- the auctioneer can cheat.
- everyone reveals their true valuations.
- it is very hard to implement.
- there is a simple dominant strategy which leads to the winner's curse.
- the seller usually refuses to accept the first price it was offered.
The Dutch auction is used to sell flowers because
- it is fast.
- it is fair.
- the Dutch like tulips.
- it is not susceptible to a winner's curse.
- it can be applied to bundles of different goods.
The Vickrey auction is a
- second-price sealed-bid auction
- second-price open-cry auction
- first-price sealed-bid auction
- first-price open-cry auction
- second-price second-bid auction
The most important thing you need to know about a Vickrey auction with private value goods is that
- the dominant strategy is to bid your true valuation
- you can win even when someone else values the good more than you
- they are very popular with business because of their real-time efficiency
- they are used in the stock market
- they suffer from the winner's curse
On which type of auction does the seller make more money,
assuming private value and risk-neutral agents?
- They are all the same
- First-price sealed-bid
Which auctions self-enforce bidder collusion agreements? (that is, cheating)
- English and Vickrey
- First-price sealed-bid and Dutch
- Dutch and Vickrey
- English and Dutch
- Vickrey and first-price sealed-bid
In a double auction, we define the surplus as
- the difference between what a buyer is willing to pay and what he actually pays.
- the payment the buyers must make to the auctioneer in order to participate.
- the number of sell bids that did not match any buy bids.
- the number of buyers divided by their average bidding price.
- the difference between the buyer's and seller's bid.
In an auction, the clearing rules specify
- who wins, what they win, and how much they pay.
- how often the auctioneer interacts with the bidders.
- in which order the bids must be placed.
- when the auctioneer must stop taking bids and sell the good.
- who must pay the auctioneer
Which one of the following is not a common
characteristic of a negotiation setting?
- A reservation price
- A set of possible proposals
- An interaction protocol
- A set of strategies for the agents
- A rule that tells when negotiation will end
Which one of the following statements about the task allocation
problem is not true?
- Each agent has a different cost function over all the subsets of tasks.
- It has a set of tasks and a set of agents.
- It has a cost allocation function which is monotonic on the number of tasks.
- The cost of doing nothing is 0.
- Initially, all the tasks are distributed among the agents.
In negotiation, a rational deal for an agent is one where
- the agent's utility either increases or stays the same.
- the agent does not have to do more tasks.
- if one agent has to do an extra task then someone else has one fewer task.
- there is no other deal where one agent receives more utility and no one receives less.
- the agent maximizes its utility.
Agents 1 and 2 are using the monotonic concession
protocol. Their negotiation will end when one of the
following happens, which one?
- The utility that 1 gets from the deal proposed by 2 is greater than the utility 1 gets from the deal it proposed to 2.
- The utility that 1 gets from the deal proposed by 2 is greater than the utility it received from the previous deal proposed by 2.
- The agents proposed deals which maximize their individual utilities (that is, 1's deal maximizes 1's utility, and the same for 2).
- Both agents receive a utility greater than 0 for the current deal.
- The time set aside for negotiations runs out.
If you are involved in negotiations with another agent that is using the Zeuthen strategy, what is your best strategy?
- The Zeuthen strategy.
- Monotonic concession.
- The Gibbard-Satterwhite strategy.
- Quit after the first round.
In the general task allocation problem with subadditve cost
function, which one of the following are viable ways for an
agent to cheat?
- Hiding tasks, creating decoy tasks, creating phantom tasks.
- Hiding tasks.
- Hiding tasks, creating decoy tasks.
- Creating decoy tasks, creating phantom tasks.
What is a pairwise comparison vote?
- A series of elections are held, on for each pair of candidates.
- Everybody votes for the two most preferred candidates.
- Everybody votes for their most preferred and least preferred candidate.
- An election is held for all candidates, then the first and second place winner go on to a pairwise election.
- Given n candidates, each voter gives his preferred candidate n votes, his next preferred candidate n-1 votes, and so on.
Which of the following desired properties of a voting system is violated by pairwise comparison?
- Rotational symmetry.
- Reflectional symmetry.
- Both rotational and reflectional symmetries.
- Reflexivity and reflectional symmetries.
Which one of the following is a problem with the Borda count?
- It is not independent of irrelevant alternatives (that is, candidates).
- It does not provide an ordering for all pairs of alternatives (that is, candidates).
- It is not Pareto efficient.
- One agent can end up being a dictator.
- The resulting ordering might not be transitive.
If using Fictitious play two agents converge on some strategy then
- this strategy is a Nash equilibrium.
- the agents will eventually diverge from it.
- this strategy is the Pareto optimal solution.
- this strategy is the social welfare solution.
- this strategy dominates all other strategies.
Under Fictitious play
- an agent remembers all the actions the other one has taken.
- an agent choses actions randomly.
- an agent pretends that it is playing against itself.
- the agent ignores the payoffs in the matrix.
- the agent pretends that the other agent is pretending to be the agent.
In the replicator dynamics for learning in games
- the agents reproduce in proportion to how much utility they have received.
- the agents replicate their opponents' past choices.
- the agents replicate each other's strategies.
- there is no possibility for convergence.
- the players with the best strategy always kill-off all the other players.
An Evolutionary Stable Strategy is defined as
- one that can overcome the presence of a small number of invaders.
- one that has no mutations.
- one that does not evolve.
- one that does well against itself.
- one that does better against all other strategies than any other strategy.
What is the moving target function problem in multiagent learning?
- As the other agents change their behavior an agent's desired behavior also changes.
- An agent is trying to catch another agent that keeps moving around in a field.
- The function that the agent is trying to learn is unknown to the agent.
- The agent's behavior influences its learning algorithm.
- The agent tries to learn the behavior function of another agent while that other agent changes its behavior.
What does the CLRI theory provide for its users?
- A formal method for determining how an agent's learning is expected to affect the learning of other agents in the system.
- A formal method for determining the optimal learning rate for an agent given a detailed description of the system and its agents.
- A reinforcement-learning algorithm for programming learning agents in multiagent systems.
- A set rules which one can use for determining which types of auctions would be better given that we are using learning agents.
- A proof that systems composed of learning agents will never converge.
What is a 1-level agent?
- One that models all other agents as state-action pairs.
- One that learns what reward it receives for each action it takes.
- One that has only one strategy.
- One that has one learning algorithm.
- One that models itself only once.
In the filtering algorithm at each turn each agent runs a
procedure called revise to two variables (its own, and
one that belongs to another agent). This procedure does the following:
- it eliminates from the agent's domain all values that cause a conflict.
- it changes the agent's value to something that does not cause a conflict.
- it asks the other agent to change its value if it conflicts with the agent's value.
- it verifies that the currently assigned values are consistent.
- it updates the agent's view.
The 4-queens problem is k-consistent for which k? Pick the largest possible k.
The hyper-resolution rule is
- a simple way to derive new no-goods from known ones.
- a way to detect constraint violations in a constraint satisfaction problem.
- a rule that states that no voting system can ever be fair.
- a rule that states that only certain facts can be deduced from a given knowledge base.
- a method for deciding which message to send to another agent in a distributed constraint satisfaction problem.
An agent that implements the asynchronous backtracking algorithm must maintain
- the list of all the nogoods it has received.
- a conversation with every agent in the system.
- the list of all the variable assignments that have been tried by the system.
- a list of the steps it has taken.
- a list of all the variable assignments it has tried for its variable.
The asynchronous backtracking is guaranteed to always finish (that is, never getting into an infinity loop) thanks to the fact that
- each agent has a different priority.
- all agents can send messages.
- only one agent can send a message at a time.
- all messages are ordered.
- all agents use the hyper-resolution rule.
The difference between asynchronous backtracking and asynchronous weak-commitment search is that
- one uses dynamic priorities and the other static.
- one is complete and the other one is not.
- one works in polynomial time while the other is exponential time.
- one of them uses backtracking but the other one does not.
- one will never find a solution, while the other one always finds a solution.
You are given a graph with nodes named a,b,c,d,e and directed
edges (a,b), (b,d), (a,c), (c,d), (b,e), (d,e). All edges have a
weight of 1. Node e is the goal node. If all agents are using
asynchronous dynamic programming and at each step all nodes do
one update, how many steps will it take for us to find the
length of the shortest path from a to e?
For the learning real-time A* algorithm, which one is
a necessary condition for the algorithm to find the answer?
- The original estimates must never overstate the true distance.
- There must not be any cycles in the graph.
- There must be one agent for each node.
- The communication mechanism between the agents must be reliable.
- The learning rate must be sufficiently large.
Problem solving using benevolent agents only is also known as
- cooperative problem solving
- task allocation
- open system engineering
- mechanism design
- a really easy problem
Distributed goal search involves using a graph that
- represents all the dependencies among goals and tasks.
- shows the exact order in which all tasks must be performed.
- links all agents with their abilities.
- shows the cost of getting from one state to another.
- shows all the constraints among variable assignments.
Why do we need to specify conventions that describe the situations
when commitments should be re-considered?
- Because if not then an agent might believe that another one has dropped a commitment.
- Because these conventions allow agents to communicate with each other.
- Because they form the basis of the contract net protocol.
- Because systems with commitments require the use of social norms.
- Because it is part of the FIPA standard.
What is a joint action commitment?
- When two agents are committed to perform a joint action.
- When two agents are committed to having one of them perform an action.
- When two agents are committed to having someone commit an action.
- When two agents commit at the same time to performing the same action.
- When two agents cannot decide who will perform an action.
Which one of the following is not a building block for achieving coordination in a multiagent system?
- social conventions
- local reasoning
Meta-level information exchange is an important coordination technique. In it
- agents send each other control level information about their priorities and goals.
- agents send each other information about the type of problem they are facing.
- agents coordinate by first deciding how they are going to achieve coordination.
- agents always tell each other everything they are doing.
- agents uses a hierarchical organizational structure to control message propagation.
Which one of the following is a concept and not just a buzzword?
- subsumption architecture
Which one of the following is a reason why you should not
use too much AI when building your industrial multiagent system?
- They will add complexity and unpredictability.
- There are no viable AI techniques available.
- Multiagent systems should never use any AI techniques.
- Machine learning only works for speech recognition.
- AI techniques will never help in building a more robust (in the face of system failures) system.
In the ADEPT project a multiagent system was built to handle
- game playing.
- social simulations.
In a typical multiagent implementation geared towards solving an
information retrieval among heterogeneous resources problem
(like the UMDL), the goal of a user agent is to
- provide the user with a unified view of the information space.
- determine how the various information resources are linked together and speed up searches.
- act as a broker between the user and other users.
- provide the user with some entertainment while he waits for the web page to download.
- save all the user preferences to a file.