Perspex Analogue of Myelin Sheaths

Mr Sam Noble, an M.Eng. student in Computer Science, has implemented a compiler for the perspex machine that is far more efficient than previous compilers and works by mimicking the way that biological neurons work.

The compiler generates artificial neurons that perform all of the actions of a computer program. In the past when a perspex program had to make a decision the neural fibres would branch, keeping a copy of the remaining part of the program on each branching fibre. This is hugely inefficient, because it implies an exponential growth in neural tissue to perform computations. Sam cut out this waste by making the compiler place "jumpers" around each neural fibre that cause the branches to jump back into segments of the main fibre. This is a major advantage for implementing the perspex machine on a standard computer.

Dr Anderson noticed that the jumpers form a sheath around a perspex fibre in much the same way as myelin forms a sheath around biological neural fibres. Myelin has a two-fold function. It speeds up the transmission of electrical signals along an axon, allowing neurons to communicate more quickly with each other, and it insulates axons so that the signals do not cross over to neurons that happen to lie close to the axon. The jumper sheaths have no effect on the speed of electrical signals in a computer, but they do prevent cross-over to nearby perspex neurons.

Dr Anderson says, "We should expect jumper sheaths and myelin sheaths to have some commonality in function, because both perspex neurons and biological neurons are geometrical structures that carry out computation."

Earlier research has found that perspex programs organise into fibres, sheets, and blocks of perspex neurons, and that synaptic gaps between perspex neurons affect computation.

Dr Anderson says, "We are seeing the emergence of a computer that mimics aspects of the biological brain for deep mathematical reasons. When I invented the perspex machine, I expected this to happen, but I am constantly surprised by the exquisite detail of the mimicry and the enormous computational advantages it brings."