Genetic Transfer or Population Diversification? Deciphering the Secret Ingredients of Evolutionary Multitask Optimization

Reading time: 6 minute
...

📝 Abstract

Evolutionary multitasking has recently emerged as a novel paradigm that enables the similarities and/or latent complementarities (if present) between distinct optimization tasks to be exploited in an autonomous manner simply by solving them together with a unified solution representation scheme. An important matter underpinning future algorithmic advancements is to develop a better understanding of the driving force behind successful multitask problem-solving. In this regard, two (seemingly disparate) ideas have been put forward, namely, (a) implicit genetic transfer as the key ingredient facilitating the exchange of high-quality genetic material across tasks, and (b) population diversification resulting in effective global search of the unified search space encompassing all tasks. In this paper, we present some empirical results that provide a clearer picture of the relationship between the two aforementioned propositions. For the numerical experiments we make use of Sudoku puzzles as case studies, mainly because of their feature that outwardly unlike puzzle statements can often have nearly identical final solutions. The experiments reveal that while on many occasions genetic transfer and population diversity may be viewed as two sides of the same coin, the wider implication of genetic transfer, as shall be shown herein, captures the true essence of evolutionary multitasking to the fullest.

💡 Analysis

Evolutionary multitasking has recently emerged as a novel paradigm that enables the similarities and/or latent complementarities (if present) between distinct optimization tasks to be exploited in an autonomous manner simply by solving them together with a unified solution representation scheme. An important matter underpinning future algorithmic advancements is to develop a better understanding of the driving force behind successful multitask problem-solving. In this regard, two (seemingly disparate) ideas have been put forward, namely, (a) implicit genetic transfer as the key ingredient facilitating the exchange of high-quality genetic material across tasks, and (b) population diversification resulting in effective global search of the unified search space encompassing all tasks. In this paper, we present some empirical results that provide a clearer picture of the relationship between the two aforementioned propositions. For the numerical experiments we make use of Sudoku puzzles as case studies, mainly because of their feature that outwardly unlike puzzle statements can often have nearly identical final solutions. The experiments reveal that while on many occasions genetic transfer and population diversity may be viewed as two sides of the same coin, the wider implication of genetic transfer, as shall be shown herein, captures the true essence of evolutionary multitasking to the fullest.

📄 Content

Abstract — Evolutionary multitasking has recently emerged as a novel paradigm that enables the similarities and/or latent complementarities (if present) between distinct optimization tasks to be exploited in an autonomous manner simply by solving them together with a unified solution representation scheme. An important matter underpinning future algorithmic advancements is to develop a better understanding of the driving force behind successful multitask problem-solving. In this regard, two (seemingly disparate) ideas have been put forward, namely, (a) implicit genetic transfer as the key ingredient facilitating the exchange of high-quality genetic material across tasks, and (b) population diversification resulting in effective global search of the unified search space encompassing all tasks. In this paper, we present some empirical results that provide a clearer picture of the relationship between the two aforementioned propositions. For the numerical experiments we make use of Sudoku puzzles as case studies, mainly because of their feature that outwardly unlike puzzle statements can often have nearly identical final solutions. The experiments reveal that while on many occasions “genetic transfer” and “population diversity” may be viewed as two sides of the same coin, the wider implication of genetic transfer, as shall be shown herein, captures the true essence of evolutionary multitasking to the fullest.

Index Terms —Evolutionary Multitasking, Genetic Transfer, Diversity, Sudoku. I. INTRODUCTION In today’s world of rapidly increasing volume, speed, and complexity of real-world challenges, the ability to effectively multitask, both cognitively as well as on computational platforms, is gaining much importance with the potential for productivity enhancement acting as the key motivation [1]. While some notable success stories can be found in the field of machine learning [2], much is yet to be explored in the context of numerical optimization and the biologically inspired search algorithms of computational intelligence. In this regard, recent works towards enhancing the population-based search algorithms of evolutionary computation (EC) for tackling multiple optimization tasks at once with a unified representation scheme, have shown considerable promise [3], [4]. In fact, it is noted that the implicit parallelism of a population is naturally well suited for the purpose of multitasking, providing the scope for spontaneous online sharing of knowledge building-blocks (in the form of encoded genetic material) across tasks. In order to further emphasize the practical significance of the above, we consider the extreme case of two minimization tasks T1 and T2 characterized by cost functions f1 and f2 of high ordinal correlation [5]. In other words, for any pair of points y1 and y2 in a unified search space Y, f1(y1) < f1(y2) ⇔ f2(y1) < f2(y2). Thus, on solving the two tasks together via multitasking, any series of search steps leading to a cost reduction of T1 automatically leads to a cost reduction of T2 for free, and vice versa, without the need for added function evaluations. As a result, it is clear that at least within the family of functions with high ordinal correlation, the idea of evolutionary multitasking (or multitask optimization) gives rise to free lunches [6], [7].
Recently in [8], a novel multifactorial evolutionary algorithm (MFEA) has been proposed as a means of exploiting the relationship between optimization tasks via the process of multitasking. The algorithm’s nomenclature follows from the fact that each task is viewed as a unique factor influencing the evolution of a single population of individuals (artificial search agents). The algorithm provides a cross- domain optimization platform and has been tested on a variety of benchmarks as well as real-world instances, involving continuous and discrete problems, often leading to noteworthy results (a summary can be found in [1]). The effectiveness of evolutionary multitasking showcased so far in the preliminary development stage has raised some interesting questions that are expected to direct future algorithmic advancements. In particular, it is considered critical to decipher the key ingredients driving the success of evolutionary multitasking. In this regard, two seemingly unrelated ideas have been put forward, namely, (a) implicit transfer of high-quality genetic building-blocks across tasks, and (b) population diversity leading to better global search.
It is noted that the boon of population diversification has been strongly emphasized in the design of island models of parallel genetic algorithms for single-tasking [9], [10], which may be seen to have some connections with the multitasking MFEA due to the implicit creation of subpopulations [8]. As a result, there is a tendency to assign the credit of MFEA’s success merely to the exploitation of population

This content is AI-processed based on ArXiv data.

Start searching

Enter keywords to search articles

↑↓
ESC
⌘K Shortcut