My research interest is in collaboration of distributed systems and
particularly of multi-agent systems (MAS). The study of MAS focuses on systems
in which many intelligent agents interact with each other. The agents are
considered to be autonomous entities, such as software programs. Cooperative
agents share a common goal, share information and collaborate to achieve the
goal. My interest is in developing architectures that facilitate the development
of multi-agent systems applications and allow collaboration, coordination and
communication between agents.
Many business environments need activity collaboration system as it increases
the efficiency of the users and makes it easier to manage information such as
meetings, contact information and documents. A
need for collaboration also exists in social environments where activity
collaboration system can improve users social interactions.
My research in multi-agent systems has two complementary themes: The possibility
to make common decisions which commit all the agents, to control the operations
that are taken by the agents, and diagnose
the faults if occur. A summary of these issues is provided below.
My main research interest is diagnosing coordination faults in distributed
systems and in MAS. In multi-agent systems agents must coordinate with each
other to achieve their goal. However, in reality this coordination sometimes
fails. The goal of my research is to make sure that coordination failures can be
detected and diagnose the reason for them. My work addresses the necessity of
diagnosing the coordination between agents as well as failures of each agent
separately. This type of diagnosis enables the multi-agent system to
automatically detect the abnormal agent(s) and its specific defective
components.
One of the main problems of diagnosis issues is the high computation and
communication complexity in large-scale multi-agent systems. Therefore, my
interest is to address this challenge and focus on techniques enabling diagnosis
of large-scale multi-agent systems by reducing the computation and communication
dramatically.
This type of diagnosis could also be implemented in distributed systems. During
the last decades, the necessity of distributed systems was increased due to the
rapid increase of users that use the systems and the physical spreading of the
systems. A central mainframe computer is not sufficient anymore in order to take
care of all the multiple missions of a system. Distributing the system to
autonomic sub-systems allows the possibility of distributing the missions
between the sub-systems. Obviously, the sub-systems should be in coordination.
My research interest is in building a diagnosis system which responsible to the
detection and diagnosis of coordination failures between the sub-systems, such
as synchronization between sub-systems, communication failures etc.
One difference between social science and computer science is that in computer
science we can formalize the system mathematically in order to be able to prove
the correctness of the algorithms as well as quantizing the efficiently of them.
Specifically in my research I use Model-Based Diagnosis (MBD), which is a formal
approach to diagnosis. In this approach the system must be formalized
mathematically so it guarantees the correctness of the algorithms and the
ability to compare between the efficiently of the algorithms.
Over the past year my research focus shifted to collaboration by voting. Voting
is a known decision making technique in complex systems in many different areas
like social science, economy etc. In MAS this technique has been adopted during
the last decade, but there are still many open issues which I wish to further
research.
First, to date, all the voting protocols assume a full connection network
between the agents. However, distributed systems could be arranged in different
topologies which may bring to generate new voting protocols.
Second, there are only a few works on voting for an incomplete set of
candidates. This situation is common in multi-agent systems where agents could
be offline, or they do not send information in order to
save communication resources.
Third, in large-scale systems the traditional voting protocols could be very
expensive in terms of communication and computation. I would like to address
this challenge and try to minimize the sent information in order to conclude
winners. In this issue, I would like to investigate the tradeoff between the
correctness of the voting protocol and their cost.