"How do you make a powerful computer with a few components and no internal wiring? Add a whiff of nitric oxide, of course."
Traditional models of neural nets only follow information flow through the neurons. Neuro-transmitters such as nitric oxide (NO) alter the way that signals are received and transmitted between neurons (neuromodulatory effects). NO is a small neuro-transmitter and can diffuse over a long area of the brain.
Researchers at the Centre for Computational Neuroscience and Robotics have constructed a simple 2D neural net that includes the effects of long-range neuro-transmitters. Their models were tested against a traditional neural net in two tasks. Both models used genetic algorithms to generate their nets.
First task: pattern recognition
Neural net: typically, 6000 generations, 46 nodes, 100 wires, 8 pixels
Gas net: typically, 1000 generations, 5-15 nodes, 2-3 pixels
"This demostrates the power provided by having two distinct yet interacting processes at play." Phil Husbands (quoted, p.39)
Second task: memory testing
Neural net: typically, 1000 generations. Gas net: typically, 100 generations.
"One curious aspect of this experiment is tha some of the final gas nets operated without any gas at all, although all of them used gas during their evolution. This suggests the gas played a key part in the learning process of the networks." (p. 39)
2. Parallels between the invisible hand in economics, genetic evolution, and genetic algorithms. (These ideas seem to be in a process of mutual justification) Gas nets and neural nets do not alter the task they are trying to learn. In economies, market agents may be considered to act in similar ways to the "neurons" in these models. (Which ways) However, market agents change the surroundings in which they operate. (Parallels: market agent, neuron; global market, neural net; geo-system resources (open to debate!), task-reward system)
In this way market systems are mutually reflexive systems. Genetic algorithms are modelled on processes of evolution in ecosystems. However, are they modelled on models of evolution in ecosystems. Population genetics has a range of models that try to mimic or predict the effect of genetic differences between organisms with an ecosystem. However, the reflexivity of an ecosystem is generally difficult to model and a wide range of, sometimes contradictory, predictions can be found.