Day 1 at IPMU'08

Field report from the Information Processing and Management of Uncertainty in Knowledge-Based Systems conference.

IPMU 08, Malaga, Spain, June 22-27, 2008.

C'est la 12e édition de la conf fondée par Bernadette Bouchon-Meunier (CNRS, LiP6) et Ronald R. Yager (Iona College, USA). ~250 participants, majorité d'Europe. Actes sur CD.

Wisdom pearls from the opening Keynote by Lofti A. Zadeh (Berkeley), father of fuzzy theory:

- A Fuzzy set is the expected value of a random set.
- Natural langage is imprecise because it is a way of describing perceptions.
- Conintention means the quality of the correspondence between the model and reality.
- Usually adjectives specialize but some adjectives generalize: A "convex set" is a special kind of set, but a set is a special kind of "random set".
- In Generalized Constraints semantics, the constraints have elasticities.
- A granular value is a value in a discretized space.

The I went to the session on reasoning:

D. Dubois on p-boxes

Generalized p-boxes are lower and upper cumulative probabilities.
A special case of evidence theory (DST, TBM...) that includes possibility.
Valuable to elicit opinion and display results (NB: besides pignistic p and pl singletons).
It's as well to stick with random sets for the number crunching in between.
Should we use "Random set" instead of "BBA" now ?

Josang on abduction

This was my first encounter with a climatosceptic. I talked with him using Beck's handbook. This makes for an uninteresting discussion, since the arguments are known in advance, and in the end the other party remains convinced that it's a conspiracy anyhow. The presentation stayed out of that debate, and the paper has a commendably readable introduction on abduction. Then it uses "subjective logic", the axioms remain to be seen and also justification of using beta probability laws. Remember to cite carefully, _only_ peer-reviewed.

Then we had a little intellectual joust:
Patrascu presentation went as "Here are the axioms of a logic with a fifth truth value." to which Dubois's presentationn answered "That is a meaningless game."

Chakraborty

I did not really understood her contribution, but at least I learned what the big word in the title meant. I am convinced that it is very important to have reasoning systems that don't explode when given contradictory inputs.

The word I learned is paraconsistency. Que faire face à la contradiction ? Dans les films, quand on programme une IA avec des ordres incohérents, ça se passe mal: voir HAL dans 2001 ou l'androide dans Alien. En logique classique, on peut démontrer n'importe quelle proposition X si "A ET non-A" est vrai. Comme les maths sont basées sur la logique classique, on espère très fort que personne n'arrivera jamais à démontrer un jour un théorème ET son contraire (et on compte espérer longtemps, la non-contradiction des mathématiques n'est pas démontrable je crois). Une façon de raisonner est paraconsistente si elle ne s'effondre pas face à la contradiction.