Title:  
Distributed Knowledge Representation in Neural-Symbolic Learning Systems

Speaker: 
Artur d'Avila Garcez

Date: 
9th June 2004

Time: 
15:20

Venue: 
Room 350

Abstract: 
Neural-symbolic systems concern the integration of the symbolic and
connectionist paradigms of Artificial Intelligence. Distributed knowledge
representation has been traditionally seen under a purely symbolic
perspective. In this talk, I'll show how neural networks can represent
symbolic distributed knowledge, acting as multi-agent systems with learning
capability (a key feature of neural networks). I'll start by showing how
simple C-ILP neural networks can be put together to encode modal logics in a
connectionist framework. I'll then apply C-ILP ensembles to solve the
well-known muddy children puzzle, a testbed for distributed knowledge
representation formalisms. A full solution to this problem requires C-ILP to
deal with knowledge evolution over time. To this end, I'll extend the
framework and show how temporal logic and combinations of temporal logics and
modal logics of knowledge can be represented in neural networks. Although
effective at representing propositional modal and temporal logics, C-ILP
ensembles lack a key feature of symbolic computation, namely, the ability to
perform recursion. I'll conclude the talk by introducing a new neural-symbolic
architecture: fibred neural networks, which is based on the idea of fibring
logical systems. Fibred networks may be composed not only of interconnected
neurons but also of other networks, forming a recursive architecture, which I
believe can represent first order logics.