The hypernetwork architecture as a model for molecular computing
aBiocomputing Lab., Wayne State University,
Detroit, MI 48202 USA
bNASA-Ames Research Center, Computational Sciences Division
This is an abstract
for a presentation given at the
Ninth
Foresight Conference on Molecular Nanotechnology.
There will be a link from here to the full article when it is
available on the web.
The fundamental challenge of nanotechnology is how to organize assemblies of molecules so that a desired function can be performed. We present a model where sets of abstract random molecules can self-organize to perform pattern recognition tasks.
The hypernetwork architecture is a biologically inspired learning model that has a representation of hierarchies [1], and is solely based on molecular interactions. Molecules are enzyme-like structures, and interactions are typical activation and inhibition processes. The representation of molecules and their interactions is comprised of binary strings and string matching respectively.
Molecules are placed in cells, modeled by cellular automata, and an organized group of cells forms an organism. Cell to cell interactions are produced by the effector-receptor molecules of the cells.
The hypernetwork receives environmental influences at its input cells, and delivers an output from its output cells. Internal cells interact with input and output cells. External influences to input cells trigger cascades of molecular interactions inside the cells of the organism.
Hypernetwork organisms learn classification tasks by an adaptive algorithm based on molecular evolution. An organism is reproduced with random molecular mutation and the selection chooses the organism with the best structure for problem to be solved [2].
This is a bottom-up approach. Molecular entities form dynamic networks of interactions, that affect the organismic level and allow an organism to perform a selected task (produce a desired output). This form of hierarchical control for molecular entities, together with molecular evolution are key elements of the hypernetwork architecture.
We propose that this model might form the basis for an approach to assembling computational devices from sets of random molecules. Just like changing synaptic weights in neural networks can produced desired input-output mappings, we show that evolving molecular structures can, in the hypernetwork model, produce equally complex input-output functions.
References:
- Segovia-Juarez, J. and Conrad, M. Hypernetwork Model of Biological
Information Processing. Proceedings of the 1999 Congress on Evolutionary Computation, July 1999, Washington DC, USA, pp: 511-515, 1999.
- Segovia-Juarez, J. and Conrad, M. Learning with the molecular-based
hypernetwork model. Proceedings of the 2001 Congress on Evolutionary
Computation, 27-30 May 2001, Seoul-Korea, pp: 1177-1182. 2001.
*Corresponding Address:
Jose Segovia-Juarez
Biocomputing Lab., Wayne State University,
4863 Second Ave. #303, Detroit, MI 48202 USA
Phone: (313)-461-5069
Fax: (313)-577-6868
Email: [email protected]
http://www.cs.wayne.edu/~jls
|