Discussion of article "Population optimization algorithms"

New comment
 

New article Population optimization algorithms has been published:

This is an introductory article on optimization algorithm (OA) classification. The article attempts to create a test stand (a set of functions), which is to be used for comparing OAs and, perhaps, identifying the most universal algorithm out of all widely known ones.

Class

When optimizing trading systems, the most exciting things are metaheuristic optimization algorithms. They do not require knowledge of the formula of the function being optimized. Their convergence to the global optimum has not been proven, but it has been experimentally established that in most cases they give a fairly good solution and this is sufficient for a number of problems.

A lot of OAs appeared as models borrowed from nature. Such models are also called behavioral, swarming or population, such as the behavior of birds in a flock (the particle swarm algorithm) or the principles of the ant colony behavior (ant algorithm).

Population algorithms involve the simultaneous handling of several options for solving the optimization problem and represent an alternative to classical algorithms based on motion trajectories whose search area has only one candidate evolving when solving the problem.

Author: Andrey Dik

 

Interesting topic - waiting for development of thoughts.

So far, finding extrema of a function is good, but is there any way to recover the function and express it mathematically, especially when optimising, say, an EA?

 
Aleksey Vyazmikin finding extrema of a function is good, but is there any way to recover the function and express it mathematically, especially when optimising, say, an EA?

1. Thank you for your interest. As there is no universally recognised ranking of algorithms, a lot of surprising discoveries are expected in the following articles: some ARs do not behave as well under detailed investigation as is commonly believed and others show extraordinary search properties. Along with classical implementations, modifications of well-known AOs will be proposed.

2. This question interests many minds of the present day, as it opens the way to such areas of knowledge as, for example, the creation of new proteins with given properties (although there is modest progress in this area, but it is achieved by simple enumeration of combinations of amino acids). In general, there are no methods of restoring the analytical formula of a function at 100%, unless only in the form of a neural network. Maybe with the application of AI technologies in the future it will be possible to reverse-engineer from data into an analytical function...

 
Andrey Dik #:

1. Thank you for your interest. As there is no universally recognised ranking of algorithms, a lot of surprising discoveries are expected in the following articles: some ARs do not behave as well under detailed investigation as it is commonly believed and others show extraordinary search properties. In addition to classical implementations, modifications of well-known AOs will be proposed.

2. This question interests many minds of the present day, as it opens the way to such areas of knowledge as, for example, the creation of new proteins with given properties (although there is modest progress in this area, but it is achieved by simple enumeration of combinations of amino acids). In general, there are no methods of restoring the analytical formula of a function at 100%, unless only in the form of a neural network. Maybe with the application of AI technologies in the future it will be possible to reverse-engineer from data into an analytical function...

Thanks for the reply.

Is there a fast method for binary variables/predictors (total volume around 5k) with gene length up to 10 letters (or whatever it's called?)?

 
Aleksey Vyazmikin #:

Thank you for your reply.

Is there a fast method for binary variables/predictors (total around 5k) with gene length up to 10 letters (or whatever it's called?)?

I don't have the answer yet, I will look for it together with the reader in future articles)))

There is a lot of research work to be done.

 
Andrey Dik #:

I don't have the answer, I will look for it together with the reader in future articles)))

There is a lot of research work to be done.

If you need to calculate something - I am ready to share the power, for the sake of science! :)

 
Aleksey Vyazmikin #:

If you need something to calculate - I'm ready to share the power, for the sake of science! :)

oh, the offer is very helpful, thanks).

 
Didn't see Bayesian optimisation in the enumeration. Or did I look too hard?
 
Vladimir Perervenko Bayesian optimisation in the enumeration. Or did you look badly?

The classification tree does not represent all existing optimisation methods to date. In addition, only population-based algorithms will be considered.

New comment