INDEX

Go To Question:

Question 1

• Why do we need automated negotiation ?
  1. Because of the distributed environment
  2. Because of limited funtionality of the agent
  3. Because all domains remain constant no matter what time constraints are given

Answer 1

Automated negotiation is needed because of the combination of distributed environments where information and tasks are shared, and limited agent funcionality, which aften results in agents interacting with one another for resources and/or information.

Answer 2

It is important in a domain in which independantly motivated agents are working because the domain attributes are going to affect the properties of the protocols. A technique that works in one domain class, motivating agents in a ceratin domain, will not neccessarily work in another type of domain.

Answer 3

The need for cooperation and negotiation arises in the Airport Resource Manager testbed because an individual agent may lack sufficient resources to satisfy its schedule and may have to borrow these resources from other agents by coordinating the individual scheduling efforts so that each agents understands the probable requirements of other agents.Then the likelihood increases that remote agents will be able to lend the appropriate resources at the time it is required.

Answer 4

A mixed deal is defined to be a division of tasks (D1,D2) with an associated probability p, agent 1 does task test D1, and with probabilty 1 2 p , it does task test set D2, (and vice versa for agent 2).This deal results in a continuous space of deals,by adding the probability into the deal definition, we make the space of deals continuous instead of discrete,the way it was originally.

Answer 5

A modular task-oriented domain is one in which the cost of combining two sets of tasks, x and y , into one large set is exactly the sum of their seperate costs minus the cost of their intersection.Of the three domains, only the fax domain is modular.

Answer 6

The idea of Reinforcement Learning is based on the idea where when an action produces a favourable result,it is reinforced.This is then applied to algorothms for learning based systems.

Answer 7

Reinforcement Learning is applied to the Iterated Prisoners Domain, where each agent can choose between two actions,cooperating or defecting.It is discovered that multiple agents' payoffs are maximised if the former strategy is applied.

Answer 8

The complete Q-learning algorithm is described as :

1. From current state s ,select an action a.
2. Update Q(s,a) based on this experience.
3. Go to 1.

Answer 9

The four difefrent properties are :

1. Autonomy
2. Social Ability
3. Responsiveness
4. Proactiveness

Answer 10

Because of the different strategies used by different agent made by different designers, the possibility of a low quality schedule of resultant negotiations is increased. So, experiments with different strategies are currently underway to solve these difficulties. Other methods,like learning agents, are also being tested out.