Source-linked AI summary
A Social Spider Algorithm for Global Optimization
James J. Q. Yu, Victor O. K. Li
TL;DR
Global optimization needs efficient search methods as real-world problems grow more complex and existing metaheuristics leave room for new methodologies. The paper proposes Social Spider Algorithm, modeling social-spider foraging through web vibrations and information sharing, and evaluates it against established algorithms. SSA shows strong benchmark performance, while the authors identify weaker competitiveness on rotated multimodal functions and preliminary parameter analysis as limitations.
Problem
Growing real-world problem complexity and the large space of possible optimization problems motivate developing new search methodologies.
Method
SSA models social-spider foraging and information sharing, using vibration source positions and intensities to guide optimization.
Results
SSA outperforms compared algorithms overall across 25 benchmark functions, with statistical advantages over CMA-ES, JADE, SaDE, and GL-25.
Takeaways & Limitations
SSA is conceptually simple, relatively easy to implement, and applicable to a wide range of continuous optimization problems.
Takeaways & Limitations
SSA is less competitive on rotated multimodal functions, and the parameter sensitivity analysis is preliminary rather than complete.
Abstract
from arXiv · showhide
The growing complexity of real-world problems has motivated computer scientists to search for efficient problem-solving methods. Metaheuristics based on evolutionary computation and swarm intelligence are outstanding examples of nature-inspired solution techniques. Inspired by the social spiders, we propose a novel Social Spider Algorithm to solve global optimization problems. This algorithm is mainly based on the foraging strategy of social spiders, utilizing the vibrations on the spider web to determine the positions of preys. Different from the previously proposed swarm intelligence algorithms, we introduce a new social animal foraging strategy model to solve optimization problems. In addition, we perform preliminary parameter sensitivity analysis for our proposed algorithm, developing guidelines for choosing the parameter values. The Social Spider Algorithm is evaluated by a series of widely-used benchmark functions, and our proposed algorithm has superior performance compared with other state-of-the-art metaheuristics.
1. Introduction
The paper motivates a new global-optimization metaheuristic from social-spider foraging, using web vibrations and information sharing to guide search. It also introduces a social foraging model, studies parameter effects, and evaluates SSA on benchmark functions.
- Growing problem size and complexity motivate efficient optimization methods based on evolutionary computation and swarm intelligence.
- Social spiders provide the basis for a new global-optimization method because they live cooperatively and detect prey through web vibrations.
- The Social Spider Algorithm models cooperative movement toward food sources by analyzing vibrations to determine promising search directions.
- The paper introduces a social animal foraging model into metaheuristic design and incorporates information-loss schemes as a distinctive algorithmic feature.
- Experiments examine how parameters and searching schemes affect SSA performance and provide inputs for further research.
- The paper evaluates SSA with benchmark functions and compares its performance with other state-of-the-art algorithms.
2. Background
Swarm-intelligence algorithms solve optimization problems by modeling collective behavior, with established approaches inspired by ants, birds, fish, bees, and fireflies. These methods encode solutions and guide search through communication, movement, or local interactions.
- Swarm intelligence models collections of interacting agents to solve difficult optimization problems.
- Ant Colony Optimization represents feasible solutions as paths and uses pheromone communication to influence ant behavior.
- Particle Swarm Optimization represents solutions as particle positions and updates movement using individual experience and social learning.
- Bee-inspired methods include algorithms that model queen mating, scout exploration, employed-bee neighborhood search, and onlooker selection.
- Glowworm Swarm Optimization uses luminescence-guided local movement, allowing subgroups to explore multiple optima.
3. Social Spider Algorithm
SSA models optimization as cooperative social-spider foraging on a hyper-dimensional web, where fitness-dependent vibrations guide movement. The paper contrasts SSA’s information-sharing and population structure with other swarm methods, then reports parameter guidance and a preliminary sensitivity-analysis limitation.
- Algorithm representation: SSA represents feasible solutions as positions on a hyper-dimensional spider web, with spiders acting as population-based optimization agents.Each spider stores its position, fitness, target vibration, movement, and dimension mask information.
- Vibration-based search: Vibrations encode fitness-dependent information that propagates across the web and are used by spiders to select guidance for movement.The vibration intensity is defined from the source position’s fitness, while propagation models attenuated information received by other spiders.
- Differences from other algorithms: SSA differs from PSO by modeling information propagation with loss and sharing current positions rather than only best-in-history positions.The authors state that the effect of information loss on optimization remains unstudied and may partially contribute to SSA’s performance over PSO.
- Differences from other algorithms: Unlike algorithms with specialized population roles, SSA uses a uniform population in which every spider performs all search tasks.The authors suggest that individuals can effectively transform among conventional role types without user guidance.
- Parameter adjustment: For multimodal problems, smaller ra and pm values are generally favored, while pc depends strongly on the optimization problem.The paper adopts ra = 1, pc = 0.7, and pm = 0.1 for later simulations, while warning that unfamiliar problems require tuning.
- Parameter adjustment: The parameter sensitivity analysis is preliminary because it varies one parameter while holding the other two fixed.The authors identify complete parameter sensitivity analysis as future research and do not guarantee the selected combination is universally best.
4. Benchmark Problems and Evaluation Method
SSA is evaluated on 25 CEC benchmark functions spanning unimodal, multimodal, rotated multimodal, and hybrid multimodal problems. The study tests multiple dimensions and compares SSA with established optimization algorithms, while omitting detailed results for several algorithms judged not comparable.
- Benchmark Functions: 25 CEC benchmark functions are organized into four groups: unimodal, multimodal, rotated multimodal, and hybrid multimodal functions.Groups I–IV are f1–f5, f6–f15, f16–f20, and f21–f25, respectively.
- Benchmark Functions: The benchmark suite tests fast convergence, escape from local optima, avoidance of premature convergence, and broader searching capability.Group I targets fast convergence; Group II contains many local minima; Group III stresses searching capability; Group IV combines components from Groups I–III.
- Experimental Settings: SSA is tested in 10, 30, and 50 dimensions, with 51 runs per function and a termination limit of 10^4 × n function evaluations.All global minima are zero, and fitness values below 10^-8 are treated as 10^-8.
- Comparison Algorithms: Comparisons include CMA-ES, JADE, SaDE, and GL-25, with source codes obtained from the original authors and minor adaptations for the benchmark functions.These algorithms were selected as state-of-the-art comparison methods, with JADE and SaDE chosen for their demonstrated CEC 2013 performance.
- Comparison Algorithms: Detailed results for Real-Coded Genetic Algorithm, Adaptive Particle Swarm Optimization, Artificial Bee Colony Optimization, Firefly Algorithm, Cuckoo Search, and Group Search Optimizer are not presented.The paper states that their overall best-fitness and convergence-speed performance was not comparable to SSA and the other four algorithms.
5. Numerical Experiments and Results
SSA is evaluated against state-of-the-art algorithms on benchmark functions across dimensions, statistical tests, convergence, reliability, and complexity. It generally performs strongly, especially on multimodal, hybrid, and higher-dimensional tests, while showing weaker convergence on unimodal and performance on rotated multimodal functions.
- Experimental Comparison: SSA generated better statistical-test results than CMA-ES, JADE, SaDE, and GL-25 on 16, 12, 14, and 13 of 25 functions, respectively.It performed no worse than those algorithms on 22, 19, 24, and 22 functions, respectively.
- Experimental Comparison: SSA achieved 7 of 10 best mean results in Group II multimodal tests, compared with 1 for CMA-ES, 4 for JADE, 1 for SaDE, and 4 for GL-25.Its three non-best mean results remained competitive and comparable to the best results.
- Experimental Comparison: SSA achieved four of five best mean results in Group IV hybrid tests, where objective-function dimensions can be unrelated without correlation-matrix information.The results identify hybrid functions as a strong test category for SSA.
- Convergence Test: SSA was generally no faster than CMA-ES, JADE, and SaDE on unimodal convergence, but it generally matched or exceeded competitors on multimodal problems.Its simultaneous exploration and exploitation can slow early convergence while helping avoid local optima.
- Reliability Test: SSA was generally more reliable than competing algorithms, with a significant advantage on 10-D and 50-D benchmarks.Its success-rate lead was similar to JADE at small thresholds but returned at larger thresholds; locating a sufficiently small region near the optimum often preceded convergence to the global optimum.
6. Conclusion
The paper proposes SSA, a social-spider-inspired algorithm for global optimization, and reports strong benchmark performance against established metaheuristics. It also identifies future directions for improving the search scheme, adapting parameters, extending the algorithm to combinatorial problems, and finding effective real-world applications.
- Conclusion: SSA is a novel algorithm based on social-spider foraging and information sharing, designed for global optimization.The authors describe it as conceptually simple, relatively easy to implement, and applicable to continuous optimization problems.
- Conclusion: SSA showed outstanding performance across three groups of 25 benchmark functions compared with CMA-ES, JADE, SaDE, and GL-25.The conclusion is supported by both simulation results and statistical analysis.
- Conclusion: Future work could improve SSA’s random-walk scheme through advanced optimization techniques, hybrid algorithms, deterministic heuristics, or local search.The authors also suggest applying new search schemes to improve performance.
- Conclusion: Future research includes adaptive parameter control, extensions to combinatorial optimization, and identifying effective real-world applications.Although SSA has only three parameters besides population size, the authors note that tuning effort could be reduced through adaptive or self-adaptive schemes.
Biography
The supplied passages identify simulation-results tables for SSA on 30-, 10-, and 50-dimensional problems.
- Biography: Table 2 presents simulation results for 30-dimensional problems.
- Biography: Table 3 presents simulation results for 10-dimensional problems.
- Biography: Table 4 presents simulation results for 50-dimensional problems.