From: Gaurav Bhaya (gbhaya@cs.washington.edu)
Date: Tue Nov 30 2004 - 21:43:50 PST
The Evolutionary origin of complex features
Richard E. Lenski, Charles Ofria, Robert T. Pennock, Christoph Adami
---------------------------------------------------------------------
One Line Summary:
- This paper verifies Darwin's theory of evolution in the context of
Digital
organisms which evolve to perform complex functions starting with simple
ones.
Important Ideas:
- The paper revolves around the main idea of evolution of digital
organisms
which consist of simple operations. The paper shows that evolution
operations
involving reproduction, mutation, inheritance, competition etc. can lead
to
organisms performing complex functions.
- The paper shows that rewards for simple functions is necessary in
order
to evolve into complex functions such as EQU which involves at least 19
operations.
Flaws:
- The paper does mention that rewards for simpler functions are
necessary
in the evolution of complex functions. However, it does not mention the
way of selecting the simpler functions. In particular, as pointed out
that
some of the simpler functions are lost in the process of obtaining the
EQU
function. The paper does not mention the selection criterion for these
reward functions. Furthermore, does the choice of these functions
matter?
- The paper uses asexual method of reproduction for generating new
members
of the population. Although it works well in this case, the paper does
not
mention and particular reason for choosing one form over another.
Moreover,
I did not understand why the population size matters if the reproduction
is asexual; i.e., how do other members of the population affect the
development of one particular gene except for the termination condition.
- The paper presents lots of number indicating various quantities such
as pivot point etc however does not elaborate on the significance of
these
quantities.
More Questions:
- How does this technique compare to existing techniques, similar ones
such as "Genetic Algorithms" or popular search methods such as simulated
annealing etc.
- What would be the effect of choosing Sexual reproduction as opposed to
asexual reproduction.
Open Questions:
- Can this idea be applied to other search problems? Does it work
as well or better than known search techniques for the problems -- If
yes,
then we have a one size fits all solution.
- How can this technique be applied to problems where solution or
"simpler
problems" are not known? How would one determine that a solution has
been
found -- In particular how do we know that a program does what it is
intended
to do.
The paper seems to discuss a new search technique based on Darwin's
theory
of evolution. The idea of rewards based on the fitness of the current
population to drive further search is used as a heuristic to guide the
search.
-- Gaurav
This archive was generated by hypermail 2.1.6 : Tue Nov 30 2004 - 21:43:55 PST