Why evolution is not a good paradigm for program induction: a critique of genetic programming
Why evolution is not a good paradigm for program induction: a critique of genetic programming
复制标题
为什么进化不是程序归纳的良好范式：对遗传编程的批判
DOI：
--
复制
发表时间：
2009
期刊：
影响因子：
--
通讯作者：
Ruibin Bai
中科院分区：
文献类型：
--
作者：
J. Woodward;Ruibin Bai
我们重新审视遗传编程的根源（即自然进化），并得出结论，进化过程的机制（即选择，遗传和变异）非常适合该过程。遗传密码是信息的有效发射器，交叉是搜索可行组合的有效方法。进化并非没有它的局限性，它可以指出，它似乎是一个非常有效的问题解决者。但是，我们过度估计了进化的问题解决能力，因为它通常试图解决“自我强加的”生存问题。我们关注图灵等效程序的演变（即具有迭代和内存的程序）。从程序归纳的角度研究了使进化效果很好的每种机制。计算机代码不像遗传代码那样强大，因此适合进化过程，导致无法有效地通过基于当前语法的基于语法的遗传操作员进行有效导航。交叉，在计算环境中采用问题，这主要是由于缺乏交换代码的上下文。对文献的评论表明，进化的程序最多包含两个嵌套环，这表明玻璃天花板可以实现目前可以完成的。
We revisit the roots of Genetic Programming (i.e. Natural Evolution), and conclude that the mechanisms of the process of evolution (i.e. selection, inheritance and variation) are highly suited to the process; genetic code is an effective transmitter of information and crossover is an effective way to search through the viable combinations. Evolution is not without its limitations, which are pointed out, and it appears to be a highly effective problem solver; however we over-estimate the problem solving ability of evolution, as it is often trying to solve "self-imposed" survival problems. We are concerned with the evolution of Turing Equivalent programs (i.e. those with iteration and memory). Each of the mechanisms which make evolution work so well are examined from the perspective of program induction. Computer code is not as robust as genetic code, and therefore poorly suited to the process of evolution, resulting in a insurmountable landscape which cannot be navigated effectively with current syntax based genetic operators. Crossover, has problems being adopted in a computational setting, primarily due to a lack of context of exchanged code. A review of the literature reveals that evolved programs contain at most two nested loops, indicating that a glass ceiling to what can currently be accomplished.