Loading...
Optimization Based on Opinion Consensus in Complex Networks
Hamed Moghadam Rafati, Homayoun | 2014
2163
Viewed
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 46667 (19)
- University: Sharif University Technology
- Department: Computer Engineering
- Advisor(s): Jalili, Mahdi
- Abstract:
- In this thesis a method based on opinion formation in complex networks aiming to solve unconstrained optimization problem has been studied. Unconstrained optimization problem is the problem of searching for the best solution in the solution space. Optimization problems have many applications in various fields that due to advances in data storage and also appearance of new big data applications solving these problems in large scales gained much importance. Different methods have been proposed in order to solve such problems but because of high computational complexity, many of them do not show good performance. A category of methods called stochastic search methods are used to overcome the problem of high computational complexity and especially a subcategory of these methods known as population based methods showed sufficient performance on many applications of these large scale optimization problems. In population based optimization methods, a population of simulated agents each corresponding to a feasible solution to the problem is used, and in order to find better solutions these feasible solutions get manipulated using a model described by mathematical equations or some predefined operations.Generally in multi-agent systems under a model of opinion formation, agents each with a vector of values called opinion vector, communicate in sequential time steps on a network that defines the connections among them. Through this communication, an interaction takes place that might change one or both agents’ opinions, using a mathematical equation. In this project, solving optimization problems using population -based method that uses the framework of opinion formation is done, based on this idea that the better (fitter) an agent’s opinion, the more its impact on others. In this study a method based on opinion formation in simulated social networks is proposed as a population-based method for binary optimization problems generally. In this propsed method other than using a novel way of performing interactions among agents, we utilized a factor called certainty on the opinion to make some sort of memory, memorizing changes taking place as time goes by. The proposed method has been compared to some widely used methods on some benchmark problems. Also a method for solving feature selection problem as a binary optimization problem is derived from the general one, using the specifications of the feature selection problem. The proposed feature selection method has been compared with two classic and a recently proposed method on some datasets. We selected datasets with higher number of dimensions to make the comparisons more valid on large-scale problems. Results of the comparisons showed that both propsed methods have superiority in performance to their rival methods
- Keywords:
- Complex Network ; Random Search ; Opinion Formation ; Multiagent System ; Binary Optimization ; Population-Based Methods ; Social Simulation ; Opinion Dynamics