My semester-long project is to create a neural network of some kind of convincing biological basis. Using very basic and simplified (yet mathematically believable) equations, I have encoded a simple microcontroller to behave like a neuron. By adjusting the parameters of this mathematical model, the computer chip can exhibit the properties of many different types of neurons. This leads to two immediate observations: I can model real biological systems, and I can create neural networks that are not biologically realizable. The first point has an application to benefit neuroscience, the other can be used in terms of artificial intelligence and networks.
It has taken me two months to build and debug the assembly code which controls an 8-bit Atmel AT90S1200 microcontroller. The code makes the chip respond to weighted stimuli in a fashion similar to that of a real neuron. This was done mostly by translating previously established neuron models to Atmel assembly (see Neuronal Modeling). The Atmel 8-bit 1200 chip was chosen because of its small size and relatively simple design and features.
Once the neuron program was debugged, I placed it on a protoboard and created a 5-cell cyclically inhibitory network. The purpose of this basic network is to demonstrate that a small number of simple neurons can produce complex behavior. In order to visualize this, I have hooked the network up to servos that act like muscles controlling a body segment of a fish swimming through water. This is what will be on display at BOOM.
The next step for my project is to build more complex, and biologically correct neural networks, and then try to apply that some basic signal processing problems and see what solutions can be found. It is exciting to see the unlimited applications of using hardware models of a very "simple" and basic component of life.