The research of the Computational Cognitive Modelling Group straddles the
overlapping and sometimes distinct areas of cortical modeling, vision research
and language research.
CORTICAL COMPUTATION RESEARCH
Dynamical systems theory is increasingly exploited as a means of understanding brain function both at a neural and cognitive level. At the neural level, a number of researchers have argued that the dynamical properties of firing neurons may have a central role to play in explaining how the brain computes. In the field of cognitive science, one of the most productive alternatives to the symbolic paradigm is the dynamical systems framework, according to which cognitive processes are behavioural patterns of non-linear dynamical systems and are best studied using the mathematics of dynamical modelling and dynamical systems theory (Port & Van Gelder, 1995; Kelso, 1997).
What is missing, however, is some attempt to bridge the gap between the neural and cognitive accounts. Given that the two levels of analysis, neural and cognitive, are now describable within the same modelling paradigm, there is an exciting opportunity to attempt to provide a unified account encompassing both levels of analysis.
- Motor cortex organization
Modeling cortical encoding of movement direction is highly relevant to the understanding of how control of movement takes place in the central nervous system and is a crucial issue in understanding and reproducing the mechanisms of visually guided reaching movements. Given the importance of the subject it is surprising that very few studies have been aimed at exploring the organisation of directional motor maps. We have designed a model which accounts for the formation of directional maps in the motor cortex and for the emergence of a population coding. Our simulation is grounded on the work of Georgopoulos and colleagues (1986) who pointed out that directional tuning is a prominent feature of motor neurons.
- Visuo-motor mapping
A fundamental operation of animal brains and robot controllers is the integration of visual information with motor commands. In our view, sensorimotor integration represents one of the most basic operation that a nervous system has to solve. We have developed a model which addresses a basic question about the visuo-motor coordination of movement: which is the mechanism that assure that visual information on the direction of movement evokes a motor response generating accurate movement in the same direction. We believe that such a simple computational mechanism allowing the correct transfer of information between a visual and a motor network might represent one of the most basic operations implemented by the central nervous system (i.e., the sensori-motor arc) and constitute evidence for the unitary nature of the sensorimotor cycle.
Ioana Marian's MSc thesis (and related papers) which addresses the above topics is downloadable from here.
- Cortical Software Re-Use (funded project)
Cortical Software Re-Use (CSRU) aims to account for a range of neural computational and cognitive development phenomena. The central concept of the theory, that of "software re-use," is borrowed from the field of software engineering. Put simply, it states that dynamical neural processes from the sensory-motor areas of the brain provide the computational building blocks for higher level functions up to and including those involved in cognition and language.
COGNITIVE LANGUAGE MODELLING and LANGUAGE RESEARCH
The language research in the Computational Cognitive Modeling Research Group
ranges widely from psycholinguistics to theoretical linguistics, all with the
ultimate focus of creating cohesive and comprehensive computational language
- DISPEL -- DIScourse Particles Expert modeL
- INTINN -- Automatic XML markup of text documents
- A Connectionist Attractor Dynamical Model of Language Development
Learning Computational Grammars (funded project)
Exploring the use of various machine learning techniques for the acquisition of grammar
CompARe: Computer Assisted Reading
With an ever-increasing amount of textual information being directed at us, it has become vitally important that we possess the necessary skills to process optimally this flood of information. Methods such as classic "speedreading" are ineffective when dealing with demanding text, or text requiring comprehension of detail (Just & Carpenter, 1987). Is it possible to facilitate and even train "optimal" reading, whereby speed is increased without diminishing comprehension? To answer this question, we need to look more closely at what exactly the eyes and the brain do during reading.
- Modelling the Reading of Thai Script
Most current models of eye movements in reading of alphabetic scripts rely on word targetting as the main strategy for guiding the eye. In alphabetic scripts that employ spaces, target selection is relatively straightforward. In the case of the Thai writing system, however, spaces are used to indicate clause boundaries (though not reliably) and ends of sentences. The lack of visible word boundaries presents the reader of Thai with a significant challenge. Central among these is how to learn reliable visual cues for locating a word's optimal viewing. It is the purpose of this project to lay the groundwork for discovering how this and other similar challenges are met. Once we have a good understanding of the low-level dynamics of eye movements in reading Thai script, it opens the way to more subtle explorations of the psycholinguistic processes involved in Thai reading. For a sample of eye movement data from a recent study carried out at the Centre for Research in Speech and Language(CRSLP), Chulalongkorn University Bangkok, Thailand, click here. This project is funded by an Enterprise Ireland International Collaboration Grant, and is a collaboration with Dr Sudaporn Luksaneeyanawin, Director of the CRSLP.
It is well known in the Computer Science Education community that students have difficulty with learning to program and this can result in high drop-out and failure rates. Identifying struggling students is difficult as introductory programming modules tend to have a very high student to lecturer ratio (100:1 or greater) and often lecturers do not know how well students are doing until after the first assessment. At this stage, it may be too late for students to withdraw from the course or to intervene to prevent struggling students from failing. This is a cause of great concern for educators and has led to a body of research in the area. Although many studies have interesting results it can be hard to know how to apply the results to other educational settings. Furthermore, the factors examined are often dependent upon the students' experience on the module and with the material and therefore it is difficult to know how predictive the factors would be if measured at the commencement of the module.
A computational model that could predict likely programming performance in the first few weeks of a module would considerably help to alleviate this problem. To build such a model would require (1) the identification of early-assessable predictors of introductory programming performance and (2) the appropriate implementation and evaluation of a scientifically sound, predictive computational model. Our research is concerned with the successful development of such a model.
This research is concerned with the development of a neuroinformatic system aimed at simplifiying and integrating the visualisation, analysis, and modelling of a range of neurotransmitter data generated from in-vivo experiments.
This project is an inter-university collaboration between the Systems Biology and Computational Cognitive Modelling Research Groups, based at NUI Maynooth, and the Microdialysis Group based in the Conway Institute of Biomolecular and Biomedical Research, University College Dublin.
The aim of this project is to develop a neuroinformatic system to simplify and integrate the visualisation and analysis of neurotransmitter release in the Motor Circuit from data derived from microdialysis in the basal ganglia of the intact, conscious rat brain. The system allows the user to:
- construct a diagrammatic abstraction of the Motor Circuit of interest;
- map experimental datasets to specific brain regions defined in the circuit schematic. Datasets are the results of typical microdialysis experiments, usually a series of neurotransmitters concentrations from a specific brain region and correlative behavioural and activity data;
- create a temporal "animation" of neurotransmitter fluctuations, providing a dynamic and quantitative description of the circuit dynamics;
- mAnipulate, compare and interact with large amounts of complex time-varying information in a coherent, visual context.