I have applied metaheuristic algorithms such as PSO, GA in my research field which is recommender system but what I have found is these algorithms are very time consuming and really not practical, though the result is better than the existing algorithms. In recommender systems, we need fast algorithm. Thank you.
Of course not !!! Metaheuristics are here to solve the problem of the time.
Metaheuristics are cleverly done since the exhaustive search with the infinite running time is useless especially for large-scale and real-life optimization problems.
Table one of the following reference gives a detailed answer to this issue:
Article Sea Lion Optimization Algorithm for Solving the Maximum Flow Problem
Dear Laishram,
Metaheuristic algorithms are optimization algorithms applied to problems which are not well defined, or have very high computational complexity. So, Metaheuristics, if applied correctly, should result in reduced computational complexity.
For example, suppose you have a function, f(x) which is not differentiable. You want to find the minimum value of the function. One method could be comparing all the values of the function corresponding to all the values of x. If the search space is very very large, it may not be a practical solution to the problem. Or, you can apply a suitable metaheuristic algorithm to find the minimum value. Metaheuristic algorithms do not search all possible values of the function to find the minimum value. So, it reduces the complexity. When the metaheuristic algorithm converses properly, should give reasonably good approximation of the desired solution. Hope, my answer makes sense.
Abdul Kayom Md Khairuzzaman Thank you for your explanation. Yes, it makes sense technically. I wanted to know whether there were any real-time applications that were really helped by these metaheuristic optimization algorithms. There are tremendous researches with these algorithms but I would be glad to find out any actual output/application of these algorithms for my knowledge (out of curiosity). For example, deep neural network in reading images used by Google in Street View. Thanks.
My hunch is that metaheuristics will fail, while modern numerical analysis tools can better take care of the data you have.
Perhaps compact algorithms will work better.
https://link.springer.com/chapter/10.1007/978-3-642-30504-7_14
Article Compact Particle Swarm Optimization
Ilya Hodashinsky Thanks for sharing Compact PSO. It is an interesting approach.
The key word here is "solve" - metaheuristics will seldom be able to find an optimum.
It should be noted that there are two main phases in recommendation methods including the training phase and test phase. In the training phase, the methods try to construct a model that will be used in the test phase to predict unknown ratings. Metaheuristic methods have been mainly used in the training phase which actually runs in offline mode. In the online mode, the unknown ratings can be predicted using the leaned models. So, the critical time consuming is about online mode (or test phase). Based on this assumption, it does not matter how much is the computational complexity of metaheuristic methods.
Does the recommender system problem have a concrete, explicit formulation as an optimization problem? If so, then it is possible to discuss the properties of the problem, and what may be the best approaches to solving it.
- Badges
- Science topic
- Similar topics
- Biological Science
- Genetics
- Genetic Analysis