Artificial Neural Networks

Daniel Simon currently works on artificial neural networks.

Neurons are the wires and processors that keep our biological circuits connected and operable. As conduits, they act to transport information (i.e. sensations, thoughts) from one place in the body to another. As processors, neurons are believed to be responsible -- and provide the mechanisms -- for such things as memory, learning, emotion, imagination, etc.

Over the past half century, the realization that a better understanding of neural signaling could lead to more efficient computational techniques -- and a deeper understanding of the mind itself -- has led researchers to put forth several models for simplified neural behavior. These models aim to idealize biological neurons as elements in a logical circuit, with the hope that a complete understanding of the simple model could help us to begin to understand the more complex biological systems. Usually, this involves treating each neuron as a non-linear summing device: a certain number of weighted inputs are summed, fed through some non-linear function (e.g. sigmoid, step-function), then output into the rest of the network.

Conducting polymers (CPs) offer ideal experimental conditions for implementing these idealized neural networks. Like biological neurons, the local concentration of ionic species controls signal propagation and conductivity. The CP-based network is simplified: the ionic species (acting as counter-ion or dopant) play an analogous role to both membrane-potential mediator and neurotransmitter in the biological network. These CP systems can also be electrochemically polymerized into two-dimensional networks with geometry roughly similar to that of cultured biological systems. Conducting polymers are readily available and robust in the experimental environment (unlike their biological analogues).

These simplified systems can then provide us with data on actualized neural-network models, and a potential arena for more efficient neural computers. They could also provide for an interface between semiconductor electronics and biological systems.

Artificial neural networks

	These are three patches of conducting polypyrrole (PPy) just before electrical connection to one another (i.e. physical contact). In order from left to right, these are configured to operate as source, gate, and drain as in a FET. In this case, however, conduction is based on polaron hopping: Thus, the system is analogous to biological conduction with synaptic weighting (i.e. the gating). The smaller circles on the bottom are 10µm across.
Confocal Microscope used to study our artificial neural networks