Tutorials & Workshop


Demystifying Nature-inspired Evolutionary Algorithms

Tutorial Details

Speaker 1

Demystifying Nature-inspired Evolutionary Algorithms

Proposer : Dr.P.Subashini, Professor of Computer Science Avinashilingam University for Women, Coimbatore, India

Conference Track : Theoretical Foundations of Artificial Intelligence

Keywords : Evolutionary Algorithms, Optimization techniques, Bio-inspired Computing, Synergistic Fibroblast Optimization, Real time applications

Abstract :

Nature Inspired Computing (NIC) paradigms are developed by inspiring natural mechanism or natural principle as source of metaphor includes evolution, ecology, development and behavior for problem solving. For the past three decades, NIC algorithms are incessantly evolved and it is considered as the best heuristic method in solving numerous complex problems. The promising solution produced by global optimization techniques motivated many researchers to utilize metaheuristic algorithms for solving the wide range of complex optimization problems in various scientific disciplines. Optimization is the most essential ingredient in the recipe of machine Intelligence. The choice of optimization algorithm can make a difference between getting a good accuracy in hours or days. The applications of optimization are limitless and is widely researched topic in industry as well as academia. In this tutorial we’ll walk through several optimization algorithms used in the realm of Machine Intelligence.

The taxonomy of NIC paradigms are partitioned into Evolutionary Computation (EC) and Swarm Intelligence (SI) techniques, in which, both are population based global optimization algorithms. Genetic Algorithm (GA) is a typical Evolutionary Algorithm (EA) which is inspired from Darwin evolution theory. On the other hand, Swarm Intelligence (SI) defines collective behavior of agents or population based natural systems including several algorithms, namely, Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), Bacterial Foraging Optimization Algorithm (BFOA), Bat Algorithm, Differential Evolution, Immune algorithms, simulating annealing, cuckoo search, culture algorithm, photosynthetic algorithm, membrane computing, intelligent water drops algorithm, harmony search and chemical – reaction ptimization have been developed. The generic algorithmic of optimization technique consists of population generation, evolution, mutation, crossover, selection and reproduction of best individuals (fittest solution) to a defined problem. Investigation on the behavior of nature inspired computing paradigms has inferred that, both evolutionary algorithms and swarm intelligence techniques shared few common characteristics, namely, population based search, natural source of inspiration, fitness function evaluation, objective to find global optimum and random based approach. The performance of global optimization techniques for solving the large scale of non-linear complex problems implied that Particle Swarm Optimization (PSO) is the most popular metaheuristic algorithm which inspired many researchers due to its characteristics, such as, cooperation, topology, searching mechanism, social behavior, interaction and mobility and the significant outcomes determines its efficiency. From the experiential analysis on the exploration and exploitation behaviors delivered by PSO algorithm and variants of PSOs, it is identified that, the premature convergence problem suffered by PSO algorithm and variant of PSOs leads to candidate solution being trapped in the local optimum (stagnation problem). It affected the movement behavior of particles to deviate from local optima and exhibit poor performance in solving highly complex problems. It reveals that introducing a new mechanism inspired by natural phenomena has significantly improved the efficiency of PSO. Henceforth, these inferential ideas laid down a foundation to acquire knowledge derived from the biological phenomena of cellular organism to introduce new parameter(s) in PSO algorithm that improves its efficiency to a great extent.

Investigation on the biological phenomena of cellular organism indicated that migration behaviour of fibroblast in the dermal wound healing process is strongly correlated with characteristics of swarm intelligence techniques. It has created an attention and passionate to further intensely learn and understand the biological phenomena and mathematical modeling of fibroblast. The intellectual behavior delivered by fibroblast in the wound healing process has motivated to carry out this research work for the construction of a novel metaheuristic algorithm. Various characteristics of fibroblast, such as, differentiation, proliferation, inflammation, migration, reorientation, alignment, extracellular matrix (ecm) synthesis, collaborative, goal-oriented, interaction, regeneration, self-adaptation and evolution, have inspired to design and develop Synergistic Fibroblast Optimization (SFO) algorithm. As an initial work, the novel biological inspired computing algorithm is tested with the standard list of benchmark test functions in order to investigate reliability, efficiency, robustness, generalization and comprehensibility to achieve the global optimum. The examined results demonstrated that the proposed synergistic fibroblast optimization algorithm is well suited with diverse characteristics of benchmark test functions involve continuous or discontinuous, differentiable or non-differentiable, separable or non-separable, scalable or non-scalable and unimodal or multimodal. It confirmed the compatibility of heuristic approach followed by SFO algorithm to find the global optima. From the observation of results, the efficiency of SFO algorithm is further validated with Congress on Evolutionary Computation (CEC) benchmark test suites and Black Box Optimization Benchmarking (BBOB) problems encompasses varied characteristics, such as, unimodal, multimodal, shifted, separable, scalable, rotated and non-separable, non-scalable, basic multimodal, composition, hybrid and composition for finding the optimal solution. The promising results signify that SFO is able to obtain best solution in solving real parameter optimization problems. Different case studies are carried out to evaluate the performance of SFO algorithm for solving complicated optimization problems, such as, combinatorial optimization problem, reinforcement learning approach and forecasting time-varying data. It is concluded that, the comparative analysis of SFO algorithm to solve benchmark problems, real world applications and case studies that comprises of both benchmark dataset and real time dataset evident that the proposed synergistic fibroblast optimization algorithm is able to achieve fairly good solution and does not get affected with premature convergence problem. The explorative analysis of results proved that SFO is competitive with other well-known global optimization algorithms and it could be considered as the best alternative metaheuristic approach for solving the non-linear complex optimization problems.

Topics to be covered : Evolutionary Computing, Developing Evolutionary Algorithms, Bio-inspired Computing, Synergistic Fibroblast Optimization, Applications of SFO in various areas of Machine Intelligence.