archives

Evolved Turing neural networks - Unorganized machines and the brain

(This is Alan Turing's 100th birthday. And I was delighted to finally stumble upon this that I had missed thus far and I find utterly interesting.)

Evolved Turing neural networks

by Craig Webster and William Fleming

In a report written in 1948 Alan Turing proposed the use of a genetic algorithm (GA) to “train” a particular type of neural network he called a B-type. Despite the apparent advantages of these networks Turing’s proposal has remained undeveloped until now [...]

Evolved design

Evolution is not a rational process and often yields solutions which are unexpected and quite different to those designed in a minimal, rational manner. As a result, evolved solutions are often very difficult to understand. For the same reason it is usually a mistake to second guess, or otherwise constrain, the “solution space” beforehand when tackling complex problems with GAs. [...]

Evolution selects only for the function of the overall network, not for the tidy compartmentalisation of function into individual components. Neural networks of all kinds and the products of GA design have the tendency to quickly become algorithmically opaque for this reason. The advantage of GAs, of course, is that evolution often works when no rational or minimal solution is known, or where finding an optimal solution by other methods would take an unreasonably long time (for example, the Travelling Salesman Problem which can take longer than the lifetime of the universe to solve).

Unorganized machines and the brain

In a fascinating and farsighted report written in 1948 Alan Turing suggested that the infant human cortex was what he called an unorganized machine[...]

Turing defined the class of unorganized machines as largely random in their initial construction, but capable of being trained to perform particular tasks. There is good reason to consider the cortex unorganized in this sense:
there is insufficient storage capacity in the DNA which controls the construction of the central nervous system to exactly specify the position and connectivity of every neurone and by not hard-wiring brain function before birth we are able to learn language and other socially important behaviors which carry great evolutionary advantage [...]

Link to the November 1946 letter of Alan Turing to William Ross Ashby, advising him to experiment with these ideas by using the ACE.

Also :

Turing's B-type neural networks and Turing's Last Programs (on AlanTuring.net), and

The Alan Turing Home Page (by Andrew Hodges)

By Cyril at 2012-06-23 01:34 | LtU Forum | 16 comments | other blogs | 7426 reads