Foundations of connectionist-symbolic integration:
representation, paradigms, and algorithms

Berlin, August 21, 2000

Background Important dates Program
  Submission procedure Organization ECAI-2000 page
  BACKGROUND In recent years much attention has been paid to the integration of connectionist systems with symbol based techniques. Whereas such an approach has clear advantages, it also encounters serious difficulties and challenges. Various models and ideas have been proposed to address various problems and aspects in this integration. The few unified approaches that have been proposed are still very limited, showing both the lack of a full understanding of the relevant aspects of this new discipline and the broad complexity in scope and tasks.

In this workshop, we aim at fostering a deep discussion about at least three topics that we believe to be fundamental for the development of a successful theory of Connectionist-Symbolic Integration: representation, paradigms, and algorithms. Concerning representation, it is fully recognized that structured representations are ubiquitous in different fields such as knowledge representation, language modeling and pattern recognition. The interest in developing connectionist architectures capable of dealing with these rich representations (as opposed to "flat" or vector-based representations) can be traced back to the end of the 80's. Today, after more than ten years since the explosion of interest in connectionism, research in architectures and algorithms for learning structured representations still has a lot to explore and no definitive answers have emerged.

Different integration paradigms have also been proposed: these are the unified and the hybrid approaches to integration. Whereas the purely connectionist ("connectionist-to-the-top") approach claims that complex symbol processing functionalities can be achieved via neural networks alone, the hybrid approach is premised on the complementarity of the two paradigms and aims at their synergistic combination in systems comprising both neural and symbolic components. In fact, these trends can be viewed as two ends of an entire spectrum.

Topics of interest include:

All these topics are usually investigated and probed independently from each other and making use of different assumptions and techniques. The organizers believe it is necessary to enforce a higher level of cross-interaction among these issues, making use of all the computational tools we have available, such as deterministic and probabilistic approaches, event-based modeling, computational logic, computational learning theory, and so on. Moreover, special attention will be given to applications domains, with the aim to devise a taxonomy that may be useful to the selection of the most suited integration paradigms and techniques to be used. We hope, also, to be able to discuss some application cases where to verify the basic ideas emerged in the literature and in the workshop's discussion itself. PARTICIPATION AND SUBMISSION PROCEDURE Participation in the workshop is open to all members of the AI community. Participants are expected to register for the main ECAI-2000 conference. The number of participants is limited. The workshop will feature invited talks, contributed presentations, and open discussion. Submitted papers will be reviewed by at least two referees. Articles reporting work in progress are encouraged. However, papers should be original and not already submitted for publication. All submissions should be sent to the organizers by e-mail, in PostScript or PDF format, to the address Common compression utilities (such as gzip, compress, or winzip) can be used. Submitted papers should not exceed 12 pages.

Other researchers interested in attending the workshop without contributing a paper should send a short position paper describing their interest in the mentioned topics.

IMPORTANT DATES Submission Deadline: March 31, 2000
Submission Notification: May 15, 2000
Final Submission Due: June 10, 2000
Workshop Held: August 21, 2000
9.00 AM R. Sun (Univ. Of Missouri, USA). (invited speaker)
Beyond Rule Extraction
9.45 AM R. Hayward, R. Nayak, and J. Diederich (Queensland Univ. of Technology, Australia).
Using Predicates to Explain Networks.
10.15 AM Discussion
10.40 AM Coffee break
11.00 AM B. Apolloni, D. Malchiodi, C. Orovas, and G. Palmas (Univ. Milano, Italy).
From Synapses to rules.
11.30 AM A.S. d'Avila Garcez, K. Broda, and D. M. Gabbay (Imperial College London, UK).
Metalevel Priorities and Neural Networks.
12.00 AM J. Neumann (Univ. Edinburgh - UK).
Holistic Transformation of Holographic Reduced Representations.
12.30 AM Discussion
12.50 AM Lunch
1.50 PM S. Kramer (Univ. Freiburg, Germany). (Invited speaker)
Decoupling Feature Construction and Model Constuction in Relational Learning.
2.35 PM Discussion
3.10 PM Coffee break
3.30 PM S. Wermter and C. Panchev (Univ. Sunderland, UK). (invited speaker)
Hybrid Sequential Machines based on Neuroscience
4.15 PM B. Hammer (Univ. Osnabrueck).
Approximation and generalization issues of recurrent networks dealing with structured data.
4.45 PM A. Wichert (Univ. Ulm, Germany).
Hierarchical Categorization in a Paleontological Research Database.
5.15 PM Discussion
Paolo Frasconi
University of Florence, Dept. Of Systems and Computer Science
Via di Santa Marta 3, I-50139 Firenze, Italy 
Marco Gori
University of Siena, Dept. Of Information Engineering
Via Roma 56, I-53100 Siena, Italy 
Franz Kurfess
Concordia University, Dept. Of Computer Science
1455 de Maisonneuve West, Montreal Quebec H3G 1M8, Canada 
Alessandro Sperduti
University of Pisa, Dept. Of Computer Science
Corso Italia 40, I-56125 Pisa, Italy