Networks and Types (Vetenskapsrådet/Swedish Research Council project VR 2013-4873) is a project led by Robin Cooper at the Department of Philosophy, Linguistics, and Theory of Science and the Centre for Language Technology at the University of Gothenburg. The project started in 2014 and will run for 3 years.
The purpose of this project is to relate types of events external to an agent (e-events) to types of events in a neural network (n-events) and to bring our work on Type Theory with Records (TTR) and Transparent Neural Networks (TNN) together in a precise way corresponding to the intuitive relationships that we have so far conjectured between them. In order to do this rigorously we will pursue three main aims.
Firstly, we will show how TTR can be used to model both e-events and n-events and an interpretation relationship between them.
Secondly, we will apply the techniques used to explore possibilities for mappings between TTR and TNN. This would provide us with a "neural interpretation" for the type theory and a "logical interpretation" for the neural nets. One reason for thinking that this is feasible is that TTR uses complex record structures for single concepts that would be represented as atoms in many standard logics. This corresponds to the fact that concepts appear to be represented in neural structure by patterns of activation rather than the activation of a single neuron. TNN is particularly suitable for this as the nets are constructed from modules which have intuitive significance.
The third aim relates to evaluating the usefulness of the mappings achieved in the first two: We will explore two related potential applications: dialogue semantics and meaning acquisition in dialogue; perceptual reasoning and meaning acquisition by robots.
Project homepage: https://sites.google.com/site/networksandtypes/home