:: ICCM 2009 ::

Tutorial Program
International Conference on Cognitive Modeling
23 July 2009

Early registration deadline: 10 June 2009



Conference home


Original call

To register, please use THIS FORM to note what tutorial(s) you will be taking. The fee for the tutorials is 10 pounds for non-students, 5 pounds for students. Payment will be taken at registration. Tutorial registration is done outside of the conference registration system.

Introduction:The Tutorials program at ICCM 2009 will be held on Thursday 23 July 2009 at the Bridgeford St. Building, at the University of Manchester. [travel directions to the campus] [map to the Bridgeford Building (#35)] The format of this year's program is modelled on previously successful ICCM tutorials, and is similar to the series held at the annual Cognitive Science Society Conference.

Registration: Tutorials cost 10 pounds for each day of tutorials for non-students and 5 pounds for students (bring ID to show at registration). You are encouraged to register through the conference site, or, if space is available, paid for on the day. Lunch can be purchased separately near the tutorial site. Attendance at the tutorials does not require conference registration; tutorial registration does not provide conference entrance. You can also register for the tutorials at the door on a space available basis, and you can have a half-day of tutorials at half-price.

There will be a meeting of the tutorial committee, tutors, and interested tutees after the tutorials; location to be announced at the tutorials.

Registration for tutorial attendees will be from 8.30 am on 23 July at the ground floor of the Bridgeford St. Building. It should take less than 5 minutes to get from the tutorial desk to the tutorial rooms, but please allow yourself this time to get to the room.

If you have a lap top, please bring it to the sessions, as you may work in pairs in several of the tutorials.

The morning session includes a 15 min. coffee break, and the afternoon session includes a 15 min. tea break.

Once in town, use the directions on the main conference site. Please note, that parking is available in fabulous, multistory carpark next to the building.



EPAM/CHREST: Fifty years of simulating learning
Lane and Gobet, Half-day (afternoon: 1345-1700)

Cognitive modeling with the neural engineering framework
Eliasmith and Stewart, Full-day (0900-1700)

Agent-based simulation: Social science simulation and beyond
Edmonds and Norling, Half-day (afternoon: 1345-1700)

A summary of Human-System Integration in the System Development Process
Ritter, Half-day (morning: 0900-1215)


EPAM/CHREST: Fifty years of simulating learning

Half-day tutorial (1345-1700) [more information on CHREST]
in Bridgeford, room to be announced

Peter Lane, University of Hertfordshire
Fernand Gobet, Brunel University

Cognitive scientists aim to develop theories which can explain and predict complex behavioural phenomena. One way of developing such theories is through detailed process models, which may be compared with data gathered from experiments. Various approaches have been taken in this area, ranging from models of isolated phenomena, such as Young and O'Shea's model of subtraction, through to integrated models such as ACT-R or Soar, and generalisable principles, such as connectionism or embodied cognition.

Within this tutorial, we present a different tradition of computational modelling, which has been providing significant models of human behaviour for 50 years. We focus on CHREST, which is derived from the EPAM (Elementary Perceiver and Memorizer) model of Feigenbaum and Simon (1984). Early models of EPAM tested the chunking theory (Chase & Simon, 1973), which has been an important component in theories of human cognition ever since. Both EPAM and CHREST build models by training a discrimination network on large quantities of natural input; CHREST also constructs templates, necessary to understand high-level expertise. Just as EPAM was the computational embodiment of the key aspects of the chunking theory (Chase & Simon, 1973), CHREST implements the essential aspects of the template theory (Gobet & Simon, 2000).

Prerequisite knowledge: We expect participants to have some general programming experience and a basic understanding of symbolic processing. Some prior knowledge of ACT-R or rule-based systems is useful but not required.

Peter Lane is a Senior Lecturer in computer science at the University of Hertfordshire, and has worked in many areas of machine and human learning, including neural networks, genetic algorithms and decision-tree learning, and in applications including language acquisition, image analysis, scientific discovery and diagrammatic reasoning. He has published more than 30 papers in academic journals and conferences.

Fernand Gobet is Professor of Cognitive Psychology at Brunel University, and closely collaborated with Herbert Simon for more than six years on the development of CHREST, the successor of EPAM. He has authored and co-authored four books, including one about modelling techniques for synthetic environments. He has written 50 refereed conference papers and over 60 refereed journal articles in leading journals.

Lane and Gobet have collaborated since 1998 on the development of CHREST and on the establishment of methodologies for developing computer models. They have presented earlier tutorials on EPAM/CHREST at international conferences on cognitive science and computational modelling.


Cognitive modeling with the neural engineering framework
Full-day tutorial (0900-1700) [more information on Neural engineering] [more on tutorial] [more on software for tutorial]
in Bridgeford Street Building, room to be announced

Chris Eliasmith
Terry Stewart
Centre for Theoretical Neuroscience, University of Waterloo

The Neural Engineering Framework (NEF; Eliasmith and Anderson, 2003) provides a general methodology for developing efficient and realistic neural models that perform a specified task. The framework consists of three quantified principles, one for each of representation, computation, and dynamics in neural systems. Adopting these principles provides a method for generating connection weights between groups of neurons that represent and transform state variables. In short, the NEF provides a neural compiler: a method for taking a high-level description of a neural system and deriving a plausible organization of realistic neurons that realize this system.

The NEF has been successfully used for modeling the barn owl auditory system, rodent navigation, zebrafish swimming control, working memory systems, and the manipulation of symbolic information. This workshop will introduce the principles of the NEF and demonstrate how they apply to cognitive modeling. This will be done through the use of Nengo, a GUI neural simulation system, which supports an adjustable level of neural accuracy, Python scripting, and the analysis of the resulting models. Extensive hands-on examples will be provided, including basic numerical processing (e.g. performing multiplication), simple motor control, working memory, and symbolic processing.

Prerequisite knowledge: We expect participants to have some general programming experience and a basic understanding of cognitive science. Participants are encouraged to bring a laptop for running Nengo, a Java/Python neural simulator that supports the Neural Engineering Framework approach. Windows, OS X, and Linux are supported equally. Participants can also share laptops among themselves.

Chris Eliasmith holds a Canada Research Chair in Theoretical Neuroscience, and is director of the new Centre for Theoretical Neuroscience at the University of Waterloo. He has over 50 publications spanning neuroscience, psychology, philosophy, computer science, and engineering, on topics including working memory, mental representation, population coding, neural dynamics, computation, automatic text classification, and cognitive architectures. His book, Neural Engineering, with Charles Anderson is now in paperback with MIT Press, and forms the basis for this workshop.

Terry Stewart is a postdoc in the Centre for Theoretical Neuroscience, after completing a PhD in Cognitive Science examining the methodological issues surrounding the creation and evaluation of computational cognitive models. His current work applies the Neural Engineering Framework to develop a neural implementation of the ACT-R production system.


Agent-based simulation: Social science simuation and beyond
Half-day tutorial (1345-1700) [more information on agent modeling]
in Bridgeford Street Building, room to be announced

Bruce Edmonds
Emma Norling
Centre for Policy Modelling, Manchester Metropolitan University

Agent-based simulation (ABS) represents each social actor with a separate encapsulated object in a simulation. The interactions between the actors are represented by the messages between these objects. These objects are usually called "agents" because it is useful to think of their attributes and internal processes in terms of what we know (or think we know) about our fellow humans. Thus, these objects usually have programs that simulate certain aspects of cognition (e.g. decision making, learning, belief representation, autonomous goals). However in a social simulation, unlike a normal cognitive model, it is those aspects that are important for social interaction that are focused on -- if you like it is the simulation you might get from a "social intelligence" stance. By the standards of cognitive models many of the programs internal to each agent might be fairly simple.

The primary application of ABS is to simulate how humans (or other social animals) might interact, for example, how complex coordination might be achieved through the interaction of essentially selfish agents. Some of these can be very complex, including many different aspects of a particular observed social situation. In this case the result if more like a dynamic description within a simulation -- a distributed representation that may incorporate many different kinds of evidence.

At the same time, complexity science has repeatedly shown how the interaction of fairly simple agents can result in complex ("emergent") outcomes. Thus, one stream of research in ABS is looking at how social systems might be understood in this way. These tend to be quite abstract simulations with very simple agents, which are intended more as a general social theory, rather than to represent any particular observed social phenomena.

When systems of independently programmed computers interact in a network, many of the same issues (trust, reputation, coordination etc.) that occur in human societies are found to be important and thus ABS can be used to investigate such situations as part of the "design" process.

This tutorial introduces the main ideas of ABS, highlighting the difficulties as well as the strength of these issues, drawing on many examples of ABS, from complex specific simulations, up to highly abstract simulations that encapsulate social theories. The first half of the tutorial looks at the more general issues and approach, whilst the second half goes through a concrete example of ABS.

Prerequisite knowledge: This tutorial is designed for post-graduate students. It assumes an interest in simulation. It should be understandable by a wide audience.

Bruce Edmonds is director of the Centre for Policy Modeling at Manchester Met. University. He has published extensively in social and socially-situated intelligence; measures and characterisations of complexity; evolutionary processes; nature and application of context in cognitive and AI domains; social simulation; philosophy of science (particularly modelling); and the application of social processes/structures to computational systems.

Emma Norling is a research associate in the Centre for Policy Modelling at Manchester Metropolitan University, working on the EPSRC Novel Approaches to Networks of Interacting Autonomes project.

Her PhD thesis is on BDI Agents for Modelling Human Behaviour.It focused on individual aspects of cognition. She is interested in social aspects of behaviour. In particular, the relationship between micro- and macro-level behaviours - in other words, the relationship between individual behaviours and society behaviour - and the complex behaviours that can arise at the macro level from simple behaviours at the micro level.


A summary of "Human-System Integration in the System Development Process"
Half-day tutorial (0900-1200) [more information on the Pew and Mavor book]
in Bridgeford Street Building, room to be announced

Frank Ritter
College of IST, Penn State

In a recent book, Pew and Mavor and the Committee for Committee on Human-System Design Support for Changing Technology (2007) propose a revision to Boehm's Spiral Model for system development. I present here a summary of this model for system design. This report argues that not understanding aspects of the user can be a risk in system design. Thus, where there are no risks, system designers do not need to worry about users. In other cases, where there are risks, the book presents approaches for reducing these risks. User models are a way to share knowledge about users across the design process.

I also include a few extensions of this model based on teaching it. These extensions are related to learning: learning by the field through using this approach to organize methods, by the system development managers learning that there are sometimes risks related to human using their systems, and by designers as lessons from one design are applied to later designs. These extensions suggest the importance of shared representations for educating team members and for the system development process. 

This tutorial will be of interest to people interested in using models in industry as a shared representation, modelers interested in applications of models, and those interested in understanding the Committee's report as edited by Pew and Mavor.

Prerequisite knowledge: This tutorial does not presume any prerequisite knowledge. Attendees may wish to have skimmed the book (which is available on the web page-at-a-time for free), or have examined other work on system design.

Frank Ritter currently teaches at the College of IST at Penn State, previously he has worked at BBN Labs and taught at the U. of Nottingham, where he was the Director of the Institute for Applied Cognitive Science. He earned his PhD in AI and Psychology and a MS in psychology from Carnegie-Mellon. He has a BSEE from the University of Illinois at Urbana/Champaign. He was a member of the committee that helped prepare the book the tutorial is based on.


Important Dates

  • 11 June 2009: Camera-ready abstract copy due for inclusion in proceedings and advertisements.
  • 29 June 2009: Camera-ready tutorial notes due (if we are to copy)

Committee members

Frank E. Ritter (Penn State, Chair)
Erik Altmann (Michigan State)
Fabio Del Missier (Trento)
Glenn Gunzelmann (Air Force Research Laboratory)
Randolph M. Jones (Soar Technology)
Katharina Scheiter (Tuebingen)
Peter Wallis (Sheffield)
Joanne Hinds (U. of Manchester)

Further contact details:

Frank Ritter
College of Information Sciences and Technology
University Park, PA 16802

Tel: + 1 814 865 4453

General Contact: www.iccm2009.net

last updated 8 June 09

Conference Overview | Venue Information | Registration | Program | Sponsors