Genetic algorithm - genetic algorithm

Japanese: 遺伝的アルゴリズム - いでんてきあるごりずむ(英語表記)genetic algorithm
Genetic algorithm - genetic algorithm

A mechanism used in evolutionary computation, inspired by the genetic mechanisms of living organisms. Calculations using genetic algorithms are called evolutionary computations, and are used in fields such as AI. The essence of the evolutionary process is trial and error, a loop of generating and selecting mutations. In other words,
(1) Generate various candidate mutations from extant species (surviving individuals).

(2) Evaluate the candidates and keep only the best ones.

(3) Return to (2) and repeat the process.

This is the same mechanism as reinforcement learning.

Mutations are generated on the basis of formal expressions such as genes in the case of living organisms, and programs in the case of evolutionary computation. These are not the actual organisms or behaviors that are created, but the expressions that become the species. This is called the genotype. Meanwhile, evaluation is performed on the grown individuals in the case of living organisms, and on the generated behaviors in the case of evolutionary computation and reinforcement learning. This is called the phenotype. It is important to note that the generation and selection of candidates are performed on different levels of objects. This is because a small change in the genotype does not necessarily correspond to a small change in the phenotype, but may result in a large change or no change. This makes it possible to jump between various phenotypes, and naturally achieves an effect similar to that of simulated annealing in neural networks (a learning method that gradually lowers the temperature (amount of change) to stabilize, like annealing metals) without falling into a local optimum.

[Hideyuki Nakajima July 19, 2019]

Source: Shogakukan Encyclopedia Nipponica About Encyclopedia Nipponica Information | Legend

Japanese:

生物の遺伝の仕組みにヒントを得てつくられた、進化計算などで用いられる仕組み。遺伝的アルゴリズムによる計算が進化計算であり、AIの分野などで利用される。進化プロセスの本質とは試行錯誤であり、変異の生成と選択のループである。すなわち、
(1)現存の種(生き残っている個体)からさまざまな変異の候補を生成する。

(2)候補を評価し、よいものだけを残す。

(3)(2)に戻りプロセスを繰り返す。

これは強化学習と同じ仕組みである。

 変異の生成は生物の場合は遺伝子、進化計算の場合はプログラムなどの形式的表現のうえで行われる。これらは実際にできあがる生物や行動そのものではなく、その種になる表現である。これを遺伝形(genotype)という。一方評価は生物の場合は成長した個体、進化計算や強化学習の場合は生成された行為のうえで行われる。これを表現形(phenotype)という。候補の生成と選択がそれぞれ別のレベルの対象で行われている点は重要である。遺伝形の小さな変化は表現形の小さな変化に対応するとは限らず、大きな変化になったり、あるいは変化しなかったりするような場合があるからである。したがって、さまざまな表現形の間をジャンプすることが可能になり、局所最適値に陥ることなく、ニューラルネットワークにおける焼きなまし法(金属の焼きなましのように、徐々に温度〔変化量〕を下げて安定させる学習手法)のような効果が自然に得られる。

[中島秀之 2019年7月19日]

出典 小学館 日本大百科全書(ニッポニカ)日本大百科全書(ニッポニカ)について 情報 | 凡例

<<:  Genetic load

>>:  Genetic counseling - Genetic counseling

Recommend

Shock Therapy

Shock Therapy 1. A type of treatment for mental di...

Rock fish

...The Scorpaenidae family is also known as the S...

Notoplana japonica (English spelling)

… [Minoru Imajima]. … *Some of the terminology th...

Soga Festival

During the Edo period, this was one of the annual ...

Tadayo Sakai - It's Sakai

A fudai daimyo in the early Edo period. Lord of M...

Kazusa Kokubun Niji - Kazusa Kokubun Niji

...former name of the province. Part of present-d...

ISSA

International Social Security Association. Founded...

Karen Blixen

A Danish female novelist. She started writing poe...

Mitsui Hachiroemon

The hereditary name that represents the Mitsui fam...

Dendrohyrax

...The gestation period is about 225 days, and th...

Photoemission

...Exoelectron emission is used in research into ...

temurah

For example, Amen AMN is an abbreviation of Adona...

Non-insulin dependent diabetes mellitus

… [Classification of disease types] Diabetes is c...

Ishizuchi Ko - Ishizuchi Ko

… [Masatane Souma] [faith] Mt. Ishizuchi has long...

Arechiha - Arechiha

…British poet, playwright, and critic. Born in St...