Home › Mark's Projects › Simple recursive systems and fractal patterns › Relationships between recursive sequences and other topics

Modeling neurons using L-Systems and genetic programming

Sun, 2007-08-12 11:22 — Mark Dow

Video of Terry Sejnowski, Salk Institute, presenting his students' computational modeling of neural form and function.

The talk is a bit slow, but the methods described are how I think computational modeling can make useful contributions to topics like growth, development and form. The goal is to describe how the variety of dendritic neural structure observed could possibly develop. There are good electro-mechanical models of neurons, but little to describe why particular dendritic trees have useful system properties, or how these trees can develop as they do. The modeling uses genetic alogithms to select for useful function, and a simple L-system to provide a range of simple forms. The resulting models are checked against properties of actual neurons to provide insights for better constraints, which are then implemented in the L-system description.

Terry Sejnowski and his group has done a bunch of interesting mathematical modeling work, including implementation of ICA (independent component analysis) algorithms and applications.

 

‹ Relationships between recursive sequences and other topicsupRelationship with human language and formal grammars ›

Comments

Mon, 2007-08-13 13:05 — PSR

L system fractals

Taking this on a tangent, part of the creativeIT grant proposal involves a fractal theory of creativity based in scale independent features of information networks. basic idea is that once a few users establish some project trees, those branching structure patterns can be generalized into a paradigm algorithm to describe the basic user-defined structuring of knowledge. then, use that minimal network algorithm to predict (potentially infinite) future project development areas and build a 'fractal' tree network of self similar individual user nodes. thats my hypothesis. should be testable. the trick will be extracting a common generalized network structure from existing user node connectivity for example from our nice growing drupal-book structures. i think L systems might be a good way of representing such an algorithm. additionally, other l-system algorithms could be used to rearrange the user networks based on different branching patterns for “assisting with breaking down rigid mental patterns” and facilitate creative associative pattern generation. I can imagine the flex RIA tree graphical user interface having some sliders to adjust l-system parameters for different info-structure branching pattern visualization.

is this description clear and if so any thoughts?


Mon, 2007-08-13 14:15 — Mark Dow

L-system network description

Yes, it seems reasonable but it needs a good graphic to demonstrate the concept. The way I read it, users have a bias toward building new nodes in stereotypical ways. Each node gives rise to x (+-y) new nodes and these nodes are related by keywords to a greater or less degree. Nodes have a probability of ending (left as leaves), and perhaps a probability of linking to other parts of the network. So it can be represented as a stochastic L-system. Do different users have different L-system parameters? I don't exactly know how this can be used predictively, but I can see how it might inform a visualization of possible future network construction.
Mon, 2007-08-13 20:36 — PSR

concept graphic

hows this for a graphic:

I pulled this from 'scale free networks'. A nice fractal. The idea is that the green-yellow cluster would represent a protypical information network on enzymind and then additional self similar subnets could be added which could be either additional user projects or the relationship between different others users information networks. yes use of networks of empty nodes for templating projects could be used for guiding futre network construction, this is a type of prediction because it would be based in the notion that if a creative association occurs in a self similar subnet in one part of the hierarchy then there is potential for similar creative events in the child networks or at relative positions at different scales of the network. wow that would be cool.

Post new comment

Security question, designed to stop automated spam bots
add Gallery2 images