Yang Lou

Analysis of the “naming game” with learning errors in communications

Naming game simulates the process of naming an objective by a population of agents organized in a certain communication network. By pair-wise iterative interactions, the population reaches consensus asymptotically. We study naming game with communication errors during pair-wise conversations, with error rates in a uniform probability distribution. First, a model of naming game with learning errors in communications (NGLE) is proposed. Then, a strategy for agents to prevent learning errors is suggested. To that end, three typical topologies of communication networks, namely random-graph, small-world and scale-free networks, are employed to investigate the effects of various learning errors. Simulation results on these models show that 1) learning errors slightly affect the convergence speed but distinctively increase the requirement for memory of each agent during lexicon propagation; 2) the maximum number of different words held by the population increases linearly as the error rate increases; 3) without applying any strategy to eliminate learning errors, there is a threshold of the learning errors which impairs the convergence. The new findings may help to better understand the role of learning errors in naming game as well as in human language development from a network science perspective.

Sequential Learnable Evolutionary Algorithm: A Research Program

Evolutionary algorithms are typically run several times in design optimization problems and the best solution taken. We propose a novel online algorithm selection framework that learns to use the best algorithm based on previous runs, hence in effect using different and better algorithms as the search progresses. First, a set of algorithms are run on a benchmark problem suite. Given a new problem, a default algorithm is run and its convergence characteristics are recorded. This is used to map to the problem database to find the most similar problem. In turn, the database returns the best algorithm for this problem and this algorithm is run in the second iteration and so on, aiming to home onto the most suitable algorithm for the problem. The resulting algorithm, named Sequential Learnable Evolutionary algorithm (SLEA), outperforms Covariance Matrix Adaptation Evolution Strategy (CMA-ES) with multi-restarts. SLEA is also applied to a new problem, a real world application, and learns its characteristics. Experimental results show that it can correctly select the best algorithm for the problem. Finally, this paper proposes a new research program which learns the algorithm-problem mapping through solving real world problems accessed through the web and worldwide cooperation through Wikipedia.

Communicating with sentences: A multi-word naming game model

Naming game simulates the process of naming an object by a single word, in which a population of communicating agents can reach global consensus asymptotically through iteratively pair-wise conversations. We propose an extension of the single-word model to a multi-word naming game (MWNG), simulating the case of describing a complex object by a sentence (multiple words). Words are defined in categories, and then organized as sentences by combining them from different categories. We refer to a formatted combination of several words as a pattern. In such an MWNG, through a pair-wise conversation, it requires the hearer to achieve consensus with the speaker with respect to both every single word in the sentence as well as the sentence pattern, so as to guarantee the correct meaning of the saying; otherwise, they fail reaching consensus in the interaction. We validate the model in three typical topologies as the underlying communication network, and employ both conventional and man-designed patterns in performing the MWNG.

Exploratory landscape analysis using algorithm based sampling

Exploratory landscape analysis techniques are widely used methods for the algorithm selection problem. The existing sampling methods for exploratory landscape analysis are usually designed to sample unbiased candidates for measuring the characteristics of the entire search space. In this paper, we discuss the limitation of the unbiased sampling and propose a novel sampling method, which is algorithm based and thus biased. Based on the sampling method, we propose several novel landscape features which are called algorithm based landscape features. The proposed features are compared with the conventional landscape features using supervised and unsupervised learning. The experimental results show that the algorithm based landscape features outperform the conventional landscape features.

Local communities obstruct global consensus: Naming game on multi-local-world networks

Abstract Community structure is essential for social communications, where individuals belonging to the same community are much more actively interacting and communicating with each other than those in different communities within the human society. Naming game, on the other hand, is a social communication model that simulates the process of learning a name of an object within a community of humans, where the individuals can generally reach global consensus asymptotically through iterative pair-wise conversations. The underlying network indicates the relationships among the individuals. In this paper, three typical topologies, namely random-graph, small-world and scale-free networks, are employed, which are embedded with the multi-local-world community structure, to study the naming game. Simulations show that (1) the convergence process to global consensus is getting slower as the community structure becomes more prominent, and eventually might fail; (2) if the inter-community connections are sufficiently dense, neither the number nor the size of the communities affects the convergence process; and (3) for different topologies with the same (or similar) average node-degree, local clustering of individuals obstruct or prohibit global consensus to take place. The results reveal the role of local communities in a global naming game in social network studies.

A Sequential Learnable Evolutionary Algorithm with a Novel Knowledge Base Generation Method

Sequential learnable evolutionary algorithm (SLEA) provides an algorithm selection framework for solving the black box continuous design optimization problems. An algorithm pool consists of set of established algorithms. A knowledge base is trained offline. SLEA uses the algorithm-problem features to select the best algorithm from the algorithm pool. Given a problem, the default algorithm is run for the initial round. After that, an algorithm-problem feature is collected and used to map to the most similar problem in the knowledge base. Then the best algorithm for solving the problem is used in the second round. This process iterates until \( n \) rounds have been made. It is revealed that the algorithm-problem feature is a good problem identifier, thus SLEA performs well on the known problems that have been encountered. However, the performance on those unknown problems is limited if the knowledge base is biased. In this paper, we propose a modified SLEA, which performs the training process using a novel method. A relatively unbiased knowledge base is formed. Experimental results show that the modified SLEA maintains the performance of SLEA on solving the CEC 2013 test suite, while it performs better than SLEA on solving a set of randomly generated max-set of Gaussian test problems.

Non-revisiting Genetic Algorithm with Constant Memory

The continuous Non-revisiting Genetic Algorithm (cNrGA) uses the entire search history and parameter-less adaptive mutation to significantly enhance search performance. Experimental results show that it has better performance than Covariance Matrix Adaptation Evolution Strategy (CMA-ES), a state of the art evolutionary algorithm. Storing the search history is natural and costs little when fitness evaluations are expensive. However, if the number of evaluations required is substantial, some memory management is desirable. In this paper, we propose two pruning mechanisms to keep the memory used constant. They are Random pruning and Least Recently Used pruning. The idea is to prune a node when a memory threshold is reached and a new node is required to be inserted, thus keeping the overall memory used constant. Experimental results show that both strategies can maintain the performance of cNrGA, up to the limit when 90% of the nodes are not recorded. This suggests that cNrGA can be extended to use in situations when the number of fitness evaluations are much larger than before with no significant effect on statistical performance, which widens the applicability of cNrGA to include more practical problems that require larger number of fitness evaluations before converging to the global optimum.

Evolving benchmark functions using kruskal-wallis test

Evolutionary algorithms are cost-effective for solving real-world optimization problems, such as NP-hard and black-box problems. Before an evolutionary algorithm can be put into real-world applications, it is desirable that the algorithm was tested on a number of benchmark problems. On the other hand, performance measure on benchmarks can reflect if the benchmark suite is representative. In this paper, benchmarks are generated based on the performance comparison among a set of established algorithms. For each algorithm, its uniquely easy (or uniquely difficult) problem instances can be generated by an evolutionary algorithm. The unique difficulty nature of a problem instance to an algorithm is ensured by the Kruskal-Wallis H-test, assisted by a hierarchical fitness assignment method. Experimental results show that an algorithm performs the best (worst) consistently on its uniquely easy (difficult) problem. The testing results are repeatable. Some possible applications of this work include: 1) to compose an alternative benchmark suite; 2) to give a novel method for accessing novel algorithms; and 3) to generate a set of meaningful training and testing problems for evolutionary algorithm selectors and portfolios.

A sequential algorithm portfolio approach for black box optimization

Abstract A large number of optimization algorithms have been proposed. However, the no free lunch (NFL) theorems inform us that no algorithm can solve all types of optimization problems. An approach, which can suggest the most suitable algorithm for different types of problems, is valuable. In this paper, we propose an approach called sequential algorithm portfolio (SAP) which belongs to the inter-disciplinary fields of algorithm portfolio and algorithm selection. It uses a pre-trained predictor to predict the most suitable algorithm and a termination mechanism to automatically stop the optimization algorithms. The SAP is easy to implement and can incorporate any optimization algorithm. We experimentally compare SAP with two state-of-the-art algorithm portfolio approaches and single optimization algorithms. The result shows that SAP is a well-performing algorithm portfolio approach.

A Bayesian Restarting Approach to Algorithm Selection.

A Bayesian algorithm selection framework for black box optimization problems is proposed. A set of benchmark problems is used for training. The performance of a set of algorithms on the problems is recorded. In the beginning, an algorithm is randomly selected to run on the given unknown problem. A Bayesian approach is used to measure the similarity between problems. The most similar problem to the given problem is selected. Then the best algorithm for solving it is suggested for the second run. The process repeats until n algorithms have been run. The best solution out of n runs is recorded. We have experimentally evaluated the property and performance of the framework. Conclusions are (1) it can identify the most similar problem efficiently; (2) it benefits from a restart mechanism. It performs better when more knowledge is learned. Thus it is a good algorithm selection framework.