Information Technology Reference
In-Depth Information
(a) Alg. I, Agent# 10, Interaction# 20
(b) Alg. II, Agent# 10, Interaction# 20
(c) Alg. I, Agent# 30, Interaction# 20
(d) Alg. II, Agent# 30, Interaction# 20
(e) Alg. I, Agent# 30, Interaction# 50
(f) Alg. II, Agent# 30, Interaction# 50
Fig. 2.
Experiments with Algorithms I and II in Different Multi-Agent Settings
on the right side of Figure 2 represent the results of these experiments. These graphs
demonstrate the same trends as algorithm I. However, more volatility is observed in the
graphs of Algorithm II compared to Algorithm I as the graphs are not monotonically
changing over the
x
axis. Indeed, this is a consequence of the increased randomization
of manipulation algorithm II compared to algorithm I.
Comparing the fusion rules, DP outperforms other fusion rules in all experiments of
Algorithm I and II. This is due to the fact that the DP rule is more categoric in its igno-
rance of the agents who are not trustworthy compared to the other two fusion rules. We
performed additional experiments and the results show that through higher number of
interactions among the agents, increase in the agents degree of trustworthiness, increase
in the number of agents in
A
, and finally decrease in the number of trust ratings (
|
T
|
),
the quality of estimation results enhances.
8
Conclusions
In this paper we analysed the properties of merging successive possibility distributions,
representing the trust of the agents in successive levels of a multi-agent system. This
