|
FLAIRS 2002 Workshop : "Causality and Categorization"
May 15, 2002, Institute for Human and Machine Cognition, Pensacola, Florida, USA
Website : http://perso.wanadoo.fr/colette.faucher/causality.html
Topic of the workshop :
This workshop is intended to further the study of the role of causal knowledge in the categorization process (how to acquire categories or concepts, use them, represent them, and so on).
More precisely, at least the following questions seem worth discussing :
- how causal information intervenes in the concept learning process,
- what is the importance of causal information to completing the usual accounts of the representation of concepts (subject-predicate, prototypical and by means of exemplars),
- how causal knowledge can be inferred from instances, to be represented within concepts, in the framework of unsupervised learning systems.
This workshop is related to the special track : "Categorization and Concept Representation : Models and Implications".
Planned format :
Four invited talks are planned (duration of each talk : 45 minutes).
Interested researchers are invited to submit papers relevant to the topic. Each presentation must not exceed 20 minutes.
Papers must be written using MS Word, RTF or PDF formats according to AAAI's standard format for authors. They must be sent to both Clark Glymour and Colette Faucher.
Only a few papers (about 6) will be presented to have time to discuss both the talks and the presented papers. Papers presented at the workshop WILL be included as workshop briefs in the published proceedings of FLAIRS. They also will be possibly published in an international journal.
The meeting will be split into a morning and an afternoon session.
Important dates :
Paper Submission Deadline : February 1, 2002
Notification of Acceptance-Rejection : February 15, 2002
Camera Ready Copy Due : March 4, 2002
Conference Date : May 15, 2002
Conditions of attendance :
Except for the invited speakers and those people presenting submitted papers, attendance will be by application only so that there can be fruitful exchange between participants in a less formal and more intimate way than during the corresponding track (see above).
Please, send your application (short CV and selected publications) to Clark Glymour and Colette Faucher.
Presentation of the invited talks :
Patricia Cheng, UCLA Department of Psychology, Los Angeles, CA, USA
Title : "Functional Basic-Level Categories and Causal Theories"
Abstract :
One might think that causal discovery depends on the definition of the entities among which causal relations are to be discovered. Categorization would therefore precede the discovery of causal relations. This paper argues for a dependence in the opposite direction: Causal discovery is the driving force underlying our mental representation of the world, not only in the sense that it is important to know how things influence each other, but also in the sense that causal relations define what should be considered things in our mental universe. The paper provides evidence that categorization does not precede causal discovery; instead, the two operate together as a single process, with optimal causal discovery being the driving force. In particular, the paper presents experiments showing that "basic-level" functional kinds are defined by causal theories of the functions in question, rather than by similarity relations. The latter are merely byproducts of the causal theories. Using identical stimuli, the experiments manipulated the level of inclusiveness of the target function to be learned. This manipulation shifted the basic level in a hierarchy, as indicated by such properties as similarity structure, verification performance, and choice of description.
Bob Rehder, Department of Psychology, New York Univ.,New York, NY, USA
Title : "A Causal-Model Theory of Conceptual Representation and Categorization"
Abstract :
This talk introduces a theory of categorization that accounts for the effects of causal knowledge that interrelates or links the features of categories. According to causal-model theory, people explicitly represent the probabilistic causal mechanisms that link category features, and classify objects by evaluating whether they were likely to have been generated by those mechanisms. Participants were taught causal knowledge that linked features of a novel category into a causal chain. In three experiments, causal-model theory provided a good quantitative account of the effect of this causal knowledge on the importance of both individual features and inter-feature correlations to classification, and did so without postulating differences in subjective feature weights or higher-order properties. By enabling precise model fits and interpretable parameter estimates causal-model theory places the "theory-based" approach to conceptual representation on equal footing with the well-known similarity-based approaches.
Steven A. Sloman, Cognitive & Linguistic Sciences, Brown Univ., Providence, RI, USA
Title : "The Psychology of Causal Reasoning"
Abstract :
I report empirical tests of a probabilistic framework for causal modeling that captures strong intuitions about human thought and reasoning, including intuitions about the nature of counterfactual reasoning and the distinction between observation and action. The fundamental claim of the framework is that people represent the world by decomposing it into autonomous mechanisms. The experiments reported examine a key assumption of the framework "its representation of actual and counterfactual intervention" in order to evaluate its viability as a source of cognitive models of categorical reasoning. The experiments focus on counterfactual inference. Implications for conceptual structure are discussed with emphasis on how people think about the functions of human artifacts.
Michael R. Waldmann, Department of Psychology, Univ. of G�ttingen, G�ttingen, GERMANY
Title : " Categories and Causal Models : A Tale of Chickens and Eggs"
Abstract :
The standard view guiding research on causality presupposes the existence of networks of causes and effects in the world that cognitive systems try to mirror. This position also underlies current research on the relationship between categories and causality. According to the view that categorization is theory-based, traditional similarity-based accounts of categorization are deficient because they ignore the fact that many categories are grounded in knowledge about causal structures. And indeed in a number of experiments, which will be summarized in the first part of the talk, we have shown that prior assumptions about the causal status of learning events governs the process of learning new categories. These studies show that learners use prior knowledge to create representations about causal models rather than associating cues with outcomes, as standard associative theories or regression models assume. However, in the second part of the talk I will present evidence that shows that the opposite direction, which thus far has largely been neglected, also holds. Categories that have been acquired in previous learning contexts may influence subsequent causal learning. It can be shown that different conceptual schemes may lead to dramatically different causal models with identical learning data. Thus, there is a bi-directional interaction between categories and causal models, similar to the relation between paradigms and scientific discovery.
Workshop co-chairs :
Clark Glymour
John Pace Scholar and Senior Research Scientist, IHMC, University of West Florida
and
Alumni University Professor of Philosophy
Department of Philosophy
Carnegie Mellon University 135 Baker Hall
Pittsburgh, PA 15213, USA
E-mail : [EMAIL PROTECTED]
Phone : (412) 268-2933
Fax : (412) 268-1440
Colette Faucher
Associate Professor of Computer Science
DIAM-IUSPIM
Facult� des Sciences de Saint-J�r�me, Universit� d'Aix-Marseille III,
Avenue Escadrille Normandie-Niemen, 13397, Marseille, Cedex 20,
FRANCE
E-mail : [EMAIL PROTECTED] and [EMAIL PROTECTED]
Phone : (+33) 4 91 05 60 58
Fax : (+33) 4 91 05 60 33
|
