Paul C. Anagnostopoulos
Nap, interrupted.
- Joined
- Aug 3, 2001
- Messages
- 19,141
Some of you may be familiar with Tom Schneider's ev program, which demonstrated information gain in a biological simulation:
http://www.lecb.ncifcrf.gov/~toms/paper/ev/
Ev simulates the co-evolution of a gene that encodes a weight matrix and a set of binding sites that are weighted by that matrix. The positions of the binding sites are predetermined, and this is what constitutes the selection pressure in the simulation. The chromosome is initialized to random bases and the simulation begins.
I.G.D. Strachan objected to the results, saying, among other things, that the predetermined sites constitutes a "fixed, pre-specified target" for the evolution. Therefore, the program is a perceptron learning application.
www.iscid.org/papers/Strachan_EvEvaluation_062803.pdf
This tempts me to conclude that evolution is nothing other than learning. The pre-specified target, rather than something specific such as pressure to locate binding sites in particular places, is the much more general pressure "survive long enough to propogate your genome." Such a pressure is so general that we wouldn't refer to it as a target, but when broken down into more specific pressures, such as "don't break apart when the temperature drops below freezing," the idea of targets is more reasonable.
~~ Paul
http://www.lecb.ncifcrf.gov/~toms/paper/ev/
Ev simulates the co-evolution of a gene that encodes a weight matrix and a set of binding sites that are weighted by that matrix. The positions of the binding sites are predetermined, and this is what constitutes the selection pressure in the simulation. The chromosome is initialized to random bases and the simulation begins.
I.G.D. Strachan objected to the results, saying, among other things, that the predetermined sites constitutes a "fixed, pre-specified target" for the evolution. Therefore, the program is a perceptron learning application.
www.iscid.org/papers/Strachan_EvEvaluation_062803.pdf
This tempts me to conclude that evolution is nothing other than learning. The pre-specified target, rather than something specific such as pressure to locate binding sites in particular places, is the much more general pressure "survive long enough to propogate your genome." Such a pressure is so general that we wouldn't refer to it as a target, but when broken down into more specific pressures, such as "don't break apart when the temperature drops below freezing," the idea of targets is more reasonable.
~~ Paul