Προβληματισμοί και κουβεντολόι..

Πολυπρακτορικά vs Κατανεμημένα Συστήματα

 
Φωτογραφία Γιάννης Παπαϊωάννου
Πολυπρακτορικά vs Κατανεμημένα Συστήματα
από Γιάννης Παπαϊωάννου - Tuesday, 23 February 2016, 8:31 PM
 

Κάθησα και έψαξα λίγο να βρω τις διαφορές των Κατανεμημένων με τα Πολυπρακτορικά Συστήματα μετά την ερώτηση του Δημήτρη και βρήκα το επισυναπτόμενο άρθρο.

 

Ουσιαστικά υπάρχουν 3 θεωρήσεις (σύμφωνα με το συγκεκριμένο άρθρο):

1) DPS (Distributed Problem Solving) is a subset of MAS (Multiagent Systems).

"That is a MAS system is a DPS system when certain assumptions hold Several such assumptions have been proposed including the benevolence assumption the common goals assumption and the centralized designer assumption."

2) MAS provides a substrate for DPS.

"While DPS thus generally assumes that whatever internal properties desired of the system can be instilled, MAS generally is concerned with how to instill these properties in the rst place. That is MAS generally only makes assumptions about the properties of individuals most typically that they are rational utility maximizers and considers what properties will emerge internally among agents, given the incentives payoffs and features of their environment."

"Thus, this view takes a divide and conquer approach to DAI (Distributed Artificial Intelligence) research. Rather than trying to jump all the way from how individual, self interested decision making on the part of each agent could lead to an overall system that accomplishes some desirable task, we can divide the problem up. MAS studies how individual, self interested decision makers might discover or be coerced into stable predictable and desirable ways of interacting among themselves. DPS then considers how these dependable desirable interactions can be initiated controlled and otherwise exploited to yield a system that accomplishes some externally defined goal."

3) MAS and DPS are complementary research agendas.

"the kinds of questionsproblems asked by MAS researchers are somewhat dierent from those asked by DPS researchers. This leads to the view that MAS and DPS are really labels not for particular kinds of systems, but rather for research agendas. As mentioned before, chances are that an observer of a system would not be able to classify it as MAS or DPS based on its observable behavior. But a system could be part of either stream of research depending on how it is experimented with."

"MAS corresponds to a research agenda that has focused on getting certain internal properties in a system of agents whose individual properties can vary. Thus, MAS has been concerned with how agents with individual preferences will interact in particular environments such that each will consent to act in a way that leads to desired global properties. MAS asks how, for a particular environment, can certain collective system properties be realized if the properties of agents can vary uncontrollably.

DPS has focused on getting external properties such as robust and efficient performance, under varying environmental conditions, from agents with established properties. DPS asks how can a particular collection of agents attain some level of collective performance if the properties of their environment are dynamic and uncontrollable."

 

Σε ένα άλλο άρθρο, αναφέρεται το εξής: "A multi-agent system (MAS) is a loosely coupled network of software entities that work together to solve problems that are beyond the individual capabilities or knowledge of each entity. A MAS is a distributed reactive system that maintains an ongoing interaction with the environment."