|
Report
1
In this
report the focus is on my personal research interest, the
role of Qualitative Reasoning (QR) in explanation generation.
Explanation
Generation & Qualitative Reasoning
At the workshop, many of the participants commented on the
need for and perceived difficulties in communicating about
models or simulation results with users, exemplified in particular
by the invited talk by Bert Bredeweg, entitled 'Communicative
Interaction and Qualitative Reasoning'. In his talk, the prospective
users are learners in an educational setting, who may develop
a model and simulate it interactively, coached by the system.
Besides tools which help the communication from system to
user, based on a 'correct model', he also indicated the need
for tools which help students to build models themselves,
and discussed techniques for interpreting student behaviour
while modelling or answering questions asked by the system.
The paper 'Application of Qualitative Process Theory to Qualitative
Simulation and Analysis of Inorganic Chemical Reaction', by
S. M. F. D. Syed Mustapha, Pang Jen-Sen and Sharifuddin Md.
Zain, also presented an educational system, based on Forbus'
QPT modelling framework. Since both the QPT framework and
Bredeweg's GARP framework use articulate and compositional
models of structure and behaviour, it seems useful to compare
the results of these projects to investigate whether it's
possible to generalize over the different techniques used,
as well as the different subject domains.
Simplification,
Aggregation, Abstraction
Because qualitative simulations can become very complex, simplifying
them is often necessary. Three papers had this as a central
theme: (1) 'Graph Theoretical Analysis of Qualitative Models
in Sustainability Science' by Klaus Eisenack and Gerhard Petschel-Held,
(2) 'Aggregation of Qualitative Simulations for Explanation'
by Anders Bouwer and Bert Bredeweg, and (3) 'Abstracting Automotive
System Models from Component-based Simulation with Multi Level
Behaviour' by Neal Snooke and Jonathan Bell. The first two
had much in common, in terms of the approach, and even some
of the techniques used to simplify state-transition graphs.
The third focused on a particular domain of electronic circuits
(in cars) and was concerned about simplifying the model of
the circuit by hiding insignificant changes based on the distinction
between system and environment variables. All three showed
how abstraction methods are related to visualization issues,
by reducing complex to simplified graphs.
Textual
explanation
Textual explanation was also addressed. Sven Kuehne and Ken
Forbus presented a poster, entitled 'Qualitative physics as
a component in natural language semantics: a preliminary report'.
Their approach is to map qualitative modelling constructs
onto (Berkeley FrameNet) frame semantics. This extends the
range of phenomena that FrameNet covers to include domains
traditionally studied in QR. Work in progress focuses on parsing
a corpus of science textbooks to support automated generation
of qualitative models.
Another poster presentation, 'Towards a framework for enhancing
Qualitative Simulation with Explanation', by Mohamed El Habib
Laraba and Zaidi Sahnoun, outlined the design goals of an
explanatory module. It should be capable of justifying each
transition, and explaining why an expected behaviour is not
present in the simulation (the second seems a much harder
problem!). Their framework includes explanatory strategies
and principles, domain knowledge (used in the simulation,
or further elaboration), dialogue history and knowledge about
effective cooperation, control knowledge, and linguistic knowledge.
Making
the right distinctions
Reflecting on the workshop as a whole, the range from technically
to conceptually oriented QR was well-represented, as well
as the range from theoretical to practical concerns. Luckily,
this did not lead to a fragmented set of presentations, but
rather to an interesting overview of issues (solutions as
well as problems), of which many seem related. For example,
explanation requires the construction of a model, but there
are many variations possible for each of these two processes,
depending on the kinds of data (numerical, textual, logical,
noise, etc.), knowledge (rules, constraints, algorithms) and
actors (QR experts, domain experts, users, learners) involved.
The common theme among all these variations in QR may well
be expressed by paraphrasing Peter Struss: 'it all comes down
to making the right distinctions'.
Anders
Bouwers, University of Amsterdam.
*****
Report
2
For me,
a young researcher that just begins my work, the participation
in the QR02 has been a very enriching experience. My small
contribution to this workshop has been a paper related with
my work. From the first moment was a pride expose my work
in this event about qualitative reasoning, they are not few
the interesting references that I remember to have found and
they have been exposed in this workshop. So, this is a good
beginning to be part of the scientific community, and better
if the exchange of knowledge is made between people that work
in the same area.
My
research area
I find the thematic of this congress the more appropriate
in order to contribute with my work. My area of interest is
centered in the application of qualitative information based
on episodes for the evaluation of dynamic systems. Our paper
presented at QR02 entitled 'Pattern Recognition Based on Episodes
and DTW. Application to Diagnosis of a Level Control System'
by J.Colomer, J.Meléndez and F.I.Gamero, is centered about
the importance of the pattern recognition in the identification
of process conditions and abnormalities.
The qualitative
representation facilitates the treatment of an enormous amount
of data available to plant operators, at the same time that
it approximates the signals to the human perception. A way
to represent qualitatively the signals is through episodes.
The numerical data conversion to qualitative data is an important
area of research in itself. Other papers presented at the
congress treats this topic, 'QSI - Qualitative Similarity
Index' by F.J. Cuberos, J.A.Ortega, R.M.Gasca and M.Toro or
'Learning Qualitative Models in the presence of noise' by
G.M.Coghill, S.M.Garrett and R.D.King.
On the
other hand, it's typical that the plant operators monitor
the state of the process looking for signs of normality or
abnormality in the process signals. Effectively, the process
conditions and abnormalities could be detected through the
patterns of the measured process data. So, a correct representation
and classification of these patterns allows users to identify
certain classes of situations.
The representation
by means of episodes of process signals applied to situation
assessment is an important support in supervisory systems.
Our contribution take in consideration the problem derived
of the comparison of patterns and show an example of application.
The future
work continues in this line, trying to create a general methodology
that could be adapted to several processes and necessities
for comparison of patterns based on episodes. The objective
is to reason through the acquired knowledge and to evaluate
the different situations of the process. In this sense, it
seems that my work will continue being related to the thematic
of the workshop.
Fco. Ignacio
Gamero, University of Girona.
*****
Report
3
My attendance
at QR02 has been a great experience. I'm surprised by the
cordial atmosphere created by everybody. The people were very
active with comments and ideas to improve or refine the work
of others. I've seen a global vision of the different applications
of QR and I've got new ideas for my research area - analysis
of temporal series from dynamic systems.
Francisco
Javier Cuberos, Systems Administrator for Radio y Televisión
de Andalucía
*****
Report
4
Preliminary
and arrival
Once we arrived in Sitges the previous day to the inaugural
sessions, a cocktail reception was offered by the organizers
which allowed all the participants to greet other participants
and to meet with the new mates. This provided a relaxed and
friendly atmosphere for the workshop which encouraged discussion
and interchange by the participants in all the sessions. It's
important to highlight that this was the general environment
at all the activities in the workshop.
Sessions
Besides the two excellent plenary sessions presented by Dr.
Bredeweg and Dr. Bratko that have been didactical of high
quality, all sessions were very interesting and all papers
were widely discussed. Several papers dealt with qualitative
simulation by using different techniques and applied to different
problems.
Another
excellent session was the posters session. In this session,
due to its characteristics, a very large discussion was established
between the posters presenters and the rest of the participants.
This is very important, because most of the posters presenters
are people who are beginning their research, and the comments
of the more experienced people are always welcome. It was
very pleasing to witness the great interest shown by consolidated
experts for the beginner's work.
The possibilities of scientific exchanges between the attendees
were extended by the different entertainment activities, which
included a guided visit to Sitges, a visit to a typical Catalan
Cava plant and, of course, the excellent gala dinner, where
everyone sample the delights of the quality of the Catalan
food.
Important
Treated Topics
As we have already said, several diverse and interesting topics
were covered. It is very important but academic background
is a very important characteristic of this workshop. There
was a high level of discussions and debate, that is to say,
the high quality of authors was interesting but the main thing
was quality of the listeners. They gave several corrections,
ideas and new research lines to authors, in a special way
to beginner authors who are doing PhDs, and therefore need
as much help as possible.
In QR
2002 were presented several works about different topics but
personally there are four topics that are interesting for
my research dynamic system, qualitative model, qualitative
simulation and order of magnitude.
We are
going to explain this because each one of these topics is
important:
We are
working on the dynamic system of Abdel-Hammid, which explains
the behaviour of software development projects. It involves
several activities, such as the assignment of effort, development,
quality, tests, measuring progress of the project, to making,
revising and modifying the initial estimation, recruiting,
training of selected staff and transfer of human resources
among projects. For these reasons, the presented papers in
QR2002 that deal with dynamic systems are interesting for
us.
Secondly,
papers about qualitative models and qualitative simulations
are interesting because in our approach, we model the dynamic
system of Abdel-Hammid in a qualitative way because the quantitative
knowledge that we have about system is little or nonexistent
most of the time. In these cases a quantitative treatment
of the system is not possible but a qualitative treatment
is possible. Qualitative simulation is especially useful when
we don't have an excess or lack of quantitative information
in order to simulate a dynamic system. Both aim to provide
a technique to carry out designs, diagnosis, analysis and
simulation where knowledge that we have of the system is little.
Qualitative
reasoning tries to incorporate the observations coming from
common sense or intuition. Furthermore, it incorporates the
expert's knowledge, and it gives qualitative explanations
of the behaviour of a system based on qualitative descriptions
of the possible situations of the real world.
The qualitative
results are sometimes a bit imprecise, but they offer a clear
idea of the possible behaviour.
Lastly,
we need define qualitative values that variables can take
to deal with the system in a qualitative way. We will simply
choose qualitative values and put them in order. We need to
define the operators of the restrictions in a qualitative
way because the values of the operands are also qualitative.
The result of the qualitative operations depends on the intervals
that we have defined and on the proportions of these intervals
in connection with the other ones. For these reasons, we think
that papers about the order of magnitude are going to be especially
interesting for our research.
Organization
It is necessary to highlight the nice organization of the
workshop because the town, Sitges, is attractive, session
room was appropriate, the hotel was comfortable, the lunch
and coffee break were delightful and the gala dinner was wonderful
and the organizers were kind and they were always willing
to solve any problem or doubt.
Antonio
J. Suárez Fábrega and Pedro J. Abad Herrera, University of
Huelva. (Spain).
*****
Report
5
I am completely
satisfied with this workshop. It is the first time that I
have attended to a conference and QR'02 has been a great experience.
The cordial atmosphere created by everybody surprises me.
I am beginning my research in the analysis of data series
and the application of these technologies in the generation
of cultural itineraries, but my work is in a preliminary step.
Sebast
Aparic, University of Seville.
*****
Report
6
During
the last QR Congress that took place in Sitges, I realised
about several things that can make me be closer to the Qualitative
Reasoning techniques in the future. The problem of dealing
with noise in the episodes based techniques is not solved
yet, but new useful approaches were proposed. At the same
time, there are other methods based on QR that can be useful
for analog electronic circuits, for example, abstracting the
envelope of a function. So, from my point of view, the QR2002
congress was very useful in that sense.
Carles
Pous, University of Girona.
*****
Report
7
This was
my first experience of an academic conference and I came away
from the event feeling that I had gained a valuable insight
into the field of Qualitative Reasoning and Model Based Systems.
It was also good to meet the people that I had been "talking"
to via email.
Janet
Thomas, MONET Project Administrator.
*****
|
