Francis Jeanson  PhD Cognitive Science: neural coding, embodied robotics,
and complex systems

Master of Science

In 2008, I completed an MSc at the University of Sussex (U.K.) in Evolutionary and Adaptive Systems (EASy). This program areas such as Artificial Life, Animal and Machine Intelligence, Neural Networks, Adaptive Systems, Computational Neural Science, Behaviour, and Development. I have specialized in computational modeling of neural mechanisms at both the structural and functional level. This includes conductance based modeling of neural activity but also reaction-diffusion modeling of chemicals, cells and networks.

Essays

Modelling Neural Maturation of the Primary Motor Cortex During Adolescence. MSc Thesis.

I begin by presenting a concise review of neural structure and function related to the primary motor cortex, realistic modeling of single neurons, oscillatory effects in the cortex, and development impacts on rhythmic activity. I then propose a computable model of a cortical neural sheet which exhibits oscillatory activity. I then explore the impact of different developmental possibilities on the coherence of synergistic activity in this sheet.

Evolving Asynchronous Cellular Automata for Density Classification. Published in Proc. of Artificial Life XI, MIT Press.

The potential of asynchronous activity in dynamical systems is investigated by evolving rules for 1D cellular automata solving the density classification problem. This research led to the discovery that asynchronous dynamics allow the genetic algorithm to find good rules more reliably than for synchronous automata. Furthermore, asynchronous forms converge more rapidly (with less computational steps) to a good solution than the synchronous forms. This suggests that asynchronous dynamics in various fields may display crucial properties that have up to here been mostly overlooked.

Rule Based modelling of Neural Development

A reaction diffusion rule based mechanism is proposed which demonstrates the strong potential of positional information theory (Wolpert 1967) to account for a large number of structural patterns present in biological organisms without the need for global information coding or sophisticated genetic regulation. The work herein further explores this for the purpose of modeling the genesis of axonal growth for neural network development.

Parsimonious modelling of Neural Plasticity

Adaptive behaviour can be explored in simple artificial agents. Here an agent that is evolved to perform phototaxis is then disrupted by inverting both its photosensors. Without further evolution this agent is then shown to adapt to this inversion by the use of two different plastic mechanisms. The first uses a recurrent neural network structure with Hebbian plasticity as inspired by DiPaolo (2000). The second uses a mass action 'experimental selection' method inspired by ideas from Edelman (1993). Finally aspects of agent pre-adaptation are explored.

Extending Embodiment: Beyond sensory-motor coordination

A theoretical criticism of current adaptive theory and method is elaborated. In particular the notion of a 'trap' from internalism in adaptive robotics is developed. An argument against a phenomenological approach for intentional behaviour is then proposed which favours the notion of 'trophism' and supports the idea of an energetic input as the basis of intentional behaviour.

Current Occupation

I currently work at the Ontario Brain Institute on the Brain-CODE neuroinformatics platform to collect, share, and analyse neuroscience research data across Ontario. More about Brain-CODE here

Doctoral Research

I completed my PhD in Cognitive Science at Carleton University in Ottawa Canada, within the Complex Adaptive Systems lab under the supervision of Dr. Anthony White.

Research Interest: Spatiotemporal coding in neural networks to understand how spike timing plays an informational role in sensorimotor integration, dynamic memory, learning, prediction and control. Because many model aspects are based on biologically realistic conditions, this work supports the idea that spatiotemporal coding play an important role in biological brains. My approach is to better understand cognition from a systems level point of view.

Philosophy: I'm a proponent of the embodied cognitive approach. I also believe a systems level view should enable us to reconciliate low-level phenomena (nerones an stuff) with high-level phenomena (having a conversation while driving) once we understand how information and dynamics are coordinated in the intermediate levels. This hierarchical view (Simon 1962) should enable a unifying approach to cognitive analysis. I believe some of these intermediate levels might be things like memory, learning, and prediction to name a few primary features as well as reward systems, biases and others as possibly fitting the role of support features... more

Previous education: Master's research within the EASy program

Academic Experience

2009-2014 PhD in Cognitive Science at Carleton University (Canada).
2007-2008 MSc in Evolutionary and Adaptive Systems at the University of Sussex (U.K.).
2001-2006 HBSc in Cognitive Science and Artificial Intelligence at the University of Toronto (Canada). Specializing in semantic modeling and cognitive models of egocentric spatial-sense acquisition.

Papers

Published

Jeanson, F., White, A. (2013). Dynamic Memory via Delay Coincidence Detection for Robot Maze Navigation. In International Conference on Cognitive Modelling (ICCM), pages 65-70.

We extend our demonstration of the possibility to harness spatiotemporal patterns in networks of coincidence detection neurones with propagation delays in a robot F-maze task in addition to new method for disambiguation between memory patterns.

Jeanson, F., Chartier, S. (2013). Memory Control in a FitzHugh-Nagumo Network via STDP. In International Conference on Cognitive Modelling (ICCM), pages 137-142.

We show how a network of Fitzhug-Nagumo cells with propagation delays can perform associative learning which can be controlled via STDP.

Jeanson, F., White, T. (2013). Dynamic Memory for Robot Control using Delay- based Coincidence Detection Neurones. In Artificial neural networks and machine learning, ICANN, Springer, pages 280-287.

Extended demonstration of robot control via spatiotemporal decoding of spiking activity in coincidence detection network of neurones in a T-maze task with context and cue stimuli.

Jeanson, F., White, A. (2013). Dynamic Memory for Robot Control via Delay Neural Networks. In Proceeding of the fifteenth annual conference companion on Genetic and evolutionary computation conference companion (GECCO 13 Companion), Christian Blum (Ed.). ACM, New York, NY, USA

First demonstration of the possibility to harness spatiotemporal patterns in networks of coincidence detection neurones with propagation delays in a robot T-maze task.

West, R., Hancock, E., Somers, S., MacDougall, K., Jeanson, F. (2013). The Macro Architecture Hypothesis: Applications to Modeling Teamwork, Conflict Resolution, and Literary Analysis. In International Conference on Cognitive Modelling (ICCM), pages 427 - 432.

In this paper we argue that although this type of work is essential, we also need explicit theory and methods for creating and evaluating cognitive models of real world tasks. This is discussed in terms of the relationship between macro and micro cognition and our own theory and methods for bridging the two.

Jeanson, F., White, A. (2012). Evolving Axonal Delay Neural Networks for Robot Control Published in Proc. of the 14th Genetic and Evolutionary Computation Conference (GECCO), 2012.

This paper investigates the dynamical and control properties of a discrete spiking neural network model with axonal delays. After examining contemporary work on spike timing as a mechanism for neural coding, we introduce a simple axonal delay network model which, via coincidence detection, demonstrates the presence of biologically observed regimes such as sustained firing and the emergence of synchrony. We establish delay criteria allowing for the classification of three distinct regimes including global synchrony, complex firing, and dissipation. We then proceed to test this model in a robot light seeking task. Results show that evolving network delays is sufficient for solving the task. We conclude by hypothesizing that global synchronous firing is more suited to reactive behaviours while complex firing patterns may serve as an organizing mechanism for more indirect processing.

Somers, S., Jeanson, F. (2011). Transductionally Bounded Hierarchical Systems. Published in Proc. of the 33rd Annual Conference of the Cognitive Science Society, 2011.

A joint paper with my colleague Sterling Somers (first author). This paper is a philosophical defense for studying cognitive systems as systems to be recognized as hierachically analyzable and bounded at the lowest level by transduction. This can be seen as an argument against the Extended Mind Hypothesis (Clark and Chalmers 1998).

Jeanson, F. (2008). Evolving Asynchronous Cellular Automata for Density Classification. Published in Proc. of Artificial Life XI, MIT Press, 2008.

The potential of asynchronous activity in dynamical systems is investigated by evolving rules for 1D cellular automata solving the density classification problem. This research led to the discovery that asynchronous dynamics allow the genetic algorithm to find good rules more reliably than for synchronous automata. Furthermore, asynchronous forms converge more rapidly (with less computational steps) to a good solution than the synchronous forms. This suggests that asynchronous dynamics in various fields may display crucial properties that have up to here been mostly overlooked.

Posters

Coincidence Detection - towards and alternative to synaptic plasticity Paper, Poster Poster at the 33rd Annual Conference of the Cognitive Science Society, 2011.

Peer reviewed poster at the 33rd annual Cognitive Science Society conference in Boston (July 2011). It's a theoretical proposal where I suggest that synaptic plasticity doesn't seem plausible as a mechanism for the very quick sensory encoding that is required for early stages of cognitive processing during echoic and working memory function. The primary argument being that the metabolic process of long term or even short term plasticity takes too much time in comparison to what our brains do in early sensory encoding stages. Instead, I propose that firing chains (which exist in virtue of coincidence detection) can, under special circumstances, function as memory stores and even build associations. Because this process would merely require the propagation of spikes, it would take place at much greater speeds than does plasticity and could therefor constitute an excellent conditate for early rapid sensory coding. This idea remains largely theoretical, yet my current work, which focusses on the formal grounding of firing chains, may soon investigate this proposal empirically via numerical simulation.

Drafts

Spatiotemporal Neural Coding Part 2 Carleton Uni., 2010

A more in depth look at spatiotemporal coding with a theoretical overview of the principles of embodied agency, evolutionary robotics, adaptation and neural coding via coincidence detection (Fujjii et al., 1996).

Spatiotemporal Neural Coding Part 1 Carleton Uni., 2009

A preliminary outlook on the field of neural coding from a temporal and spatial point of view. This served as a the first 'proof of concept' for my doctoral work.

Enactive Perceptual Supplementation Carleton Uni., 2009

A proposal and investigation of how the 'enactive approach' (Varela et al., 1990) can serve as a framework to guide the design of perceptual supplementation devices for both dissabled individuals and within high cognitive load scenarios.

Philosophy continued... Intelligence enables adaptation, of which we not only do but do so actively. We don't simply react intelligently to events but predict what might occur under existing conditions and also under hypoethetical ones (simulated conditions). We then act upon these in order to satisfy some internal conditions (Ashby 1956). Rather than a substance view of representational theory of mind, I favour a process view. We can speak of cognition as the act of building internal models of the external world (learing), using them to anticipate the environment and others (prediction), and finally controlling ones, mental process, body action, or other objects surrounding (control). Perhaps the greatest predictive achievement in biological systems is the ability for sensorimotor and symbolic integration which leads to an advanced form of sense-making (`understanding'). Despite some of their faults, the Chinese Room and Symbol Grounding Problem are classic whistleblowers of classical AI (Searle 1980, Harnad 1987). By basing our theoretical investigation of intelligence in embodiment, adaptation, complexity, computation and dynamics a unified theory of cognition is in reach.

Technical

I'm a software developer when I'm not reading or writing. I write software for my simulation work but also for web and desktop applications. I enjoy programming because I find it allows a kind of artistic expression via formal syntax. My general interests with regard to programming are quite broad and I usually get excited with just about any project...

I've coded frameworks (two at the moment) that aren't fully featured (or stable) that I might release someday, written in php, with the idea that the end user (with some programming knowledge) can develop a web site with minimal code.

I've also developed a neural controlled 2D agent simulator with evolution application that is cross-platform compatible (thanks to Qt); which might also get released onto the web someday.

I like working with Python for the fast drafting of ideas and web-apps but still hold C as my prefered language for quick run-time results during simulations.

Professional Experience

2009-2009 Web Developer for Gap Adventures. Developing dynamically served web pages, forms, and applications in Django Python and JQuery. SEO adviser for the deployment of the new web site.
2006-2007 Software Developer for Metavera Solutions Inc. Developing Object Oriented web based applications.
2004-2006 Software Developer for Quur. Developing desktop and web applications. e.g. Interactive conversation agents, web content management systems, SQL database systems, Flash applications, AJAX applications.

Areas of Expertise

Research: Cognitive Modeling, Biological Neural Modeling, Evolutionary Algorithms, Complexity Theory, Dynamical Systems, Logic, Natural Language Processing, Semantics.
Programming: C, C++, Python, Matlab, Java, PHP, Prolog, LISP
Frameworks: MVC (Model View Controller), Qt4, SDL, ODE (Physics modeling).
Databases: MySQL and PosgreSQL on Apache.
Graphics: 3D OpenGL, Flash, Photoshop, Final Cut Pro, After Effects.
Web: CMS (Joomla, Droopal, WordPress), SEO, Prototype, JQuery.
Languages: Perfectly bilingual in French and English written and spoken.

Contact Info

Tel: +1 613 769-2209 (Canada)
Email: fjeanson(at)yahoo.com


Last update: January 2015 - Francis Jeanson