About
my research
My
research interests are in the broad areas of
Artificial Intelligence (AI) and Multi-Agent
Systems (MAS). During my PhD and in the following
years I have been working on intelligent
autonomous agents reasoning and coordination,
logic and
argumentation-based dialogue and negotiation,
specification and
verification of agent interaction. In
particular, I have been working on runtime
monitoring and exception handling in
interaction protocols, developing modeling
frameworks to specify agent interaction using
declarative tools such as backward and
forward rules, the event
calculus, and social
commitments. Recent advances in norms
for agent societies make me believe that much of
the work done on interaction protocols and
commitments may have a value also in the context
of contract-based and normative multi-agent
systems.
I
did my best to promote Computational Logic and MAS
research through the activities of the Italian
Association for Logic Programming
(GULP),
with my involvement in the steering committees of
the DALT
and CLIMA
workshops series, and by organizing the ISCL
and DALT
Spring Schools.
In
these last years, I have been fascinated by the
emerging field of computational social
sciences, where I believe that a great
deal of progress made in logical reasoning
frameworks, especially
argumentation-based, and multi-agent
systems, can find promising application
and have a positive impact on the society. I also
believe that computer science research, such as
that discussed at AAMAS, COMMA, IJCAI conferences
and the like, has a lot to learn from other
disciplines such as social and cognitive sciences,
and philosophy (but it seems reluctant to admit
it).
I
am always open to collaborations. If you are
interested in these topics and want to share
ideas, feel free to drop me a line. I also
encourage you to try out the software
described in some of my publications:
The ALIAS architecture
for distributed abductive reasoning,
whereby agents can share hypotheses and reach
conclusions in a collaborative fashion, using
partial knowledge;
The SCIFF framework ,
for interaction protocol specification and
verification. The SCIFF language is very simple
and yet powerful: you can use it to specify the
way you wish a system to behave, and then you
can use the SCIFF proof-procedure to monitor at
runtime whether the system does follow your
specifications. You can express knowledge bases.
SCIFF can be used in contexts such as
multi-agent interaction and normative
systems. SCIFF has been a joint
endeavour between the AI group at the University
of Bologna, and colleagues at the University of
Ferrara (see below).
CLIMB,
a specialization of SCIFF, somehow tailored to
web service choreographies and business
protocol specifications. CLIMB drew
inspiration from work by Pesic and van der Aalst
on Delare and ConDec. CLIMB's conceptual
framework and implementation were mainly due to
Marco
Montali, currently in Bolzano;
jREC,
a Java runtime monitor for open systems
specified using REC, a reactive version
of the Event Calculus, particularly
apt to tracking fluents
online. jREC has been implemented
by Marco Montali;
ComMon,
a runtime monitor for multi-agent
commitments. The commitment
specification language, REC, allows you to
specify properties that refer to a knowledge
base, and reason with metric time, i.e., "real"
deadlines (not just "next" or "until" and the
like). That is why you should check it out if
you want to develop practical applications
(i.e., for diagnosing
exceptions in multi-agent contracts) and
you wouldn't mind doing that based on
solid theoretical background. jREC has
been implemented by Marco Montali;
NetArg,
a NetLogo model for agent based social
simulations, in particular for
studying opinion dynamics [check
out this video]. NetArg agents are
empowered with abstract argumentation frameworks
and are meant to strike a good compromise
between more basic, very common representation
of opinions (e.g., an array of real values) and
more complicated, but less popular BDI-like
representations of cognitive agents. With NetArg
we want to
model not only the dynamics of opinions in
social network, but also the reasons behind
opinions, expressed as arguments. NetArg
was mainly implemented by Simone Gabbriellini;
TwitterArg,
a NetLogo model for analyzing microdebates,
i.e., streams of Tweets about a certain topic,
tagged so as to mark opinions and relations
between opinions. Arguments are built bottom-up
around opinions, by grouping together Tweets
about the same opinion. TwittterArg uses ConArg,
a CP-based Java reasoner developed by Stefano
Bistarelli and Francesco Santini, to compute the
semantics of abstract argumentation frameworks
in an efficient way. Arguments may have
different weights, based on the number of Tweets
or re-Tweets. NetArg
was mainly implemented by Simone Gabbriellini.
The
rest of this page mainly discusses my research
activities up to 2006. For my more recent
activities, check my
publications, and those by my
co-authors and by other
authors interested in related topics.
Logic-based
Multi-Agent Systems
MAS
are communities of problem-solving entities that
can perceive and act upon the environment to
achieve their individual goals as well as joint
goals. Work on such systems integrates many
technologies and concepts in AI and other areas
of computing as well as other disciplines. Over
recent years, the agent paradigm gained
popularity, due to its applicability to a full
spectrum of domains, from search engines to aids
to electronic commerce and trade, e-procurement,
recommendation systems, distributed diagnosis,
simulation and routing, knowledge management and
distributed systems configuration. In such
domains, a centralized approach based on a
single solver in charge of managing all aspects
of a distributed computation would be
inadequate. A multi-agent approach, instead,
seems to be more promising in that it would
allow for collaboration of multiple solvers,
which will coordinate with each other and
exhibit autonomy of reasoning and decision
making in solving relevant sub-tasks.
Although
commonly implemented by means of imperative
languages, mainly for reasons of efficiency,
agent-related concepts have recently increased
their influence in the research and development of
computational logic-based systems [AMAI
2004]. Computational logic provides a
general framework for studying syntax, semantics
and procedures for agents. Besides, it lends
itself to the specification and verification agent
interaction, and to the implementation and
description of environments, tools, and standards
for MAS.
In
the context of logic-based MAS, my research
activity has focussed on three main aspects:
coordination of agent reasoning, agent
negotiation, and agent interaction.
Co-ordination
of agent reasoning
One
of the main engineering principles of MAS is
locality. Each agent is responsible for operations
within a limited domain, and will base most of his
reasoning on domain-specific knowledge. However,
when many agents interact with each other, each
one operating based on its own knowledge and
beliefs, it may become necessary for agents to
take into account information and results produced
by other agents. It is then necessary to extend
classical reasoning techniques so as to enable
agents to cope with multiple and incomplete
knowledge, and to interact with each other, for
information sharing and possibly for consistency
maintenance purposes. In this context, abduction
is a well known hypothetical reasoning mechanism
in AI aimed at explaining observations or guessing
causes of effects, based on an incomplete
knowledge.
Agents
characterised by exhibiting abductive reasoning
capabilities are called abductive agents. Since my
doctoral research activity started [PhD
2002], I have been investigating different
interaction patterns for abductive agents. I have
been working on extending existing abductive logic
programming-based proof-procedures towards
multi-agent reasoning and on the application of
the developed techniques to different scenarios.
Among the main results achieved in this direction
are the Abductive LogIc AgentS architecture
(ALIAS) presented in [AMAI
2003], and the language LAILA, defined in [AIxIA
Notizie 2000,CompLang
2001] to coordinate the reasoning activity
of multiple abductive agents. I have contributed
to the implementation of ALIAS and LAILA [ALIAS,AIxIA
1999], and to their testing in several
application domains, including negotiation over
resources [AMAI
2003], distributed diagnosis [CLIMA
2000], recommendation systems [COCL
1999,MAS-LP
1999], judicial evaluation of criminal
evidence [AAIJ
2004]. Finally, I have contributed to
extending ALIAS/LAILA towards coordination of
constraint-based reasoning in agent systems [JELIA
2002a,AAMAS
2002a].
Logic
and argumentation-based negotiation
Negotiation
for resource achievement is one of the most
extensively studied areas in multi-agent research.
It has been approached from many perspectives,
including game theory, auction theory, and more
recently argumentation. Many agent architectures
and frameworks for negotiation are only
theoretical studies or practical implementations.
The
main aims of my work on logic- and
argumentation-based negotiation have been to
provide a well-specified, general purpose agent
framework, with a rigorous operational semantics
defined to bridge the gap between specification
and implementation. Following Kowalski and Sadri's
pioneering work, "From Logic Programming
Towards Multi-Agent Systems" [AMAI
1999], my colleagues and I have identified
abduction as the main form of reasoning to
reconcile knowledge coming from the outside of
agents (through negotiation) with internal goals
and needs.
I
pursued this work mainly in collaboration with the
Agents and Logic Programming groups of Imperial
College London and of the University of Cyprus. I
started working on the definition of an
argumentation-based agent dialogue framework [AISB
2001] during a research visit to Imperial
College London in 2000. We understood that most
available argumentation-based negotiation proposed
abstract frameworks, in which agents were not
implemented but only represented by their
knowledge bases. Protocols were mostly implemented
by data structures based on state automata.
Considering abduction as an inference mechanism
able to support argumentation processes, we
proposed a new approach based on abductive logic
agents [ATAL
2001], which gained a best paper award at a
prestigious event (ATAL). Since 2001, such a
framework has been refined and extended in several
directions, and presented at the main AI, logics,
and agent conferences [ECAI
2004,AAMAS
2002b,IJCAI
2003,JELIA
2002b] and workshops [ESAW
2001,UKMAS
2002a]. Based on the seminal work of 2000
and 2001, I have actively contributed to the
preparation and writing of a European project
proposal, which has been accepted in Fall 2001.
From 2002 to 2005, my research activity has mainly
focussed on the goals of the EU-funded "Societies
Of ComputeeS" project [SOCS],
especially for what concerns specification and
verification of agent interaction.
Agent
interaction: specification and verification
Agent
Communication Languages (ACL) and Interaction
Protocols (IP) are generally defined for
heterogeneous agents to effectively co-operate
with each other. The study of ACL and IP
represents nowadays a major research direction in
MAS, and they are well known for being a well
suited domain for formal approaches. The two major
schools of thought in this area tend to interpret
agent communication either from a motivational and
causal perspective, or from a social and open
perspective, The former aims to link the semantics
of agent interaction to some agent architecture,
grounded on the notion of mental states, whereas
the latter seeks to keep such semantics
independent of the agents' internals, and "open"
to its application to societies of heterogeneous
autonomous agents. Following this second main
stream, the SOCS social model developed within the
SOCS project, to which I have actively
contributed, has been firstly presented [UKMAS
2002b] in association with deontic
categories of social obligation and commitment.
During
the course of the SOCS project, I have worked on
the definition of the SOCS social model in terms
of declarative semantics, initially with a deontic
flavour [CEEMAS
2003,FAMAS
2003], and subsequently with an abductive
interpretation [AIxIA
2003], and in terms of operational semantics
[AIxIA
2005,GC
2004,CLIMA
VI,DEIS-LIA-06-001].
I have contributed towards the implementation of
the SOCS social model, called the "SOCS Social
Infrastructure" (SOCS-SI) [SOCS,AAIJ
2006,AT2AI
2004,AAMAS
2004]. Finally, I have contributed to
investigating the application of SOCS-SI to a
range of domains. Among them, I have considered
protocols for resource sharing [DALT
2003], combinatorial auction protocols [CISI
2004,Intelligenza
Artificiale 2005], protocols for electronic
transactions [SAC
2004] and other protocols including
human-to-human e-mail-based IP and security
protocols [ESAW
2005]. Recently my group and I provided a
mapping of the SOCS social model onto deontic and
normative frameworks [CMOT
2006], we extended the SOCS-SI framework
towards automated verification of agent
interaction [ENTCS
2003], and we started the development of a
methodology for protocol design whose main driver
is property achievement [WETICE
2004].
Other
work
Alongside
my research on Logic-Based MAS, I have also
contributed to the area of generative and reactive
planning [ISMIS
2000,UK
PS-SIG 1999] in collaboration with the
constraint programming group of DEIS. Together
with colleagues, I have investigated the
combination of logics and meta-heuristics for
multi-agent resource allocation [STAIRS
2002]. Finally, I have worked on the
implementation of abductive proof-procedures using
Constraint Logic Programming [AGP
2003,CILC
2004], and to the definition of new models
of agent interaction and information sharing [CLIMA
IVa ,CLIMA
IVb].
Impact
of my research
According
to Google
Scholar, in April 2013 my H-Index is 22, my
i10-Index is 42. Here
there is an incomplete list of articles citing my
work.
|
|
|