Yintian Liu

A novel method for real parameter optimization based on Gene Expression Programming

Gene Expression Programming (GEP) is a new technique of evolutionary algorithm that implements genome/phoneme representation in computing programs. Due to its power in global search, it is widely applied in symbolic regression. However, little work has been done to apply it to real parameter optimization yet. This paper proposes a real parameter optimization method named Uniform-Constants based GEP (UC-GEP). In UC-GEP, the constant domain directly participates in the evolution. Our research conducted extensive experiments over nine benchmark functions from the IEEE Congress on Evolutionary Computation 2005 and compared the results to three other algorithms namely Meta-Constants based GEP (MC-GEP), Meta-Uniform-Constants based GEP (MUC-GEP), and the Floating Point Genetic Algorithm (FP-GA). For simplicity, all GEP methods adopt a one-tier index gene structure. The results demonstrate the optimal performance of our UC-GEP in solving multimodal problems and show that at least one GEP method outperforms FP-GA on all test functions with higher computational complexity.

The Strategies of Initial Diversity and Dynamic Mutation Rate for Gene Expression Programming

Gene expression programming (GEP) is a genotype/phenotype system that evolves candidate solutions encoded in linear chromosomes of fixed length. Its characteristics make GEP surpass other evolutionary algorithms. However, the original GEP algorithm generates the initial population in a simple way and uses a constant mutation rate determined empirically. Two effective strategies are proposed to overcome these limitations. The contributions of this paper include: (1) developing an algorithm to extract the Open Reading Frame of a gene without parsing the Expression Tree, (2) defining the consanguineous relation between chromosomes, (3) proposing an algorithm to diversify the initial population based on the consanguineous relation, and (4) proposing a novel adaptive mutation rate strategy for each chromosome in evolution. Furthermore, a performance evaluation using both synthetic and real-life data demonstrates that the proposed strategies are effective. The result shows that the correlation coefficient between observed values and predicted values can be increased by as high as 0.1856, and the classification problem can be solved in less running time.

Network Intrusion Active Defense Model Based on Artificial Immune System

Based on artificial immune theory, a new model of active defense for analyzing the network intrusion is presented. Dynamically evaluative equations for self, antigen, immune tolerance, mature-lymphocyte lifecycle and immune memory are presented. The concepts and formal definitions of immune cells are given, the hierarchical and distributed management framework of the proposed model are built. Furthermore, the idea of biology immunity is applied for enhancing the self-adapting and self-learning ability to adapt continuously variety environments. The experimental results show that the proposed model has the features of real-time processing, self-adaptively, and diversity, thus providing a good solution for network surveillance.

A Comparative Study of Correlation Measurements for Searching Similar Tags

In recent years, folksonomy becomes a hot topic in many research fields such as complex systems, information retrieval, and recommending systems. It is essential to study the semantic relationships among tags in folksonomy applications. The main contributions of this paper includes: (a) proposes a general framework for the analysis of the semantic relationships among tags based on their co-occurrence. (b)investigates eight correlation measurements from various fields; then appliying these measurements to searching similar tags for a given tag on datasets from del.icio.us. (c) conducts a comparative study on both accuracy and time performance of the eight measurements. From the comparison, a best overall correlation measurement is concluded for similar tags searching in the applications of folksonomy.

Application of Gene Expression Programming to Real Parameter Optimization

Gene Expression Programming (GEP) is a new evolutionary algorithm that implements genome/phoneme representations. Despite its powerful global search ability and wide application in symbolic regression, little work has been done to apply it to real parameter optimization. A real parameter optimization method named Uniform-Constant based GEP (UC-GEP) is proposed in this paper. The main work and contributions include: (1) Compares UC-GEP with Meta-Constant based GEP (MC-GEP), Meta-Uniform-Constant based GEP (MUC-GEP), and Floating Point Genetic Algorithm (FP-GA) on optimizing seven benchmark functions, respectively. Experiment results show that GEP methods outperform FP-GA on five of the seven functions and UC-GEP reaches the global optimum on all seven functions. (2) Compares UC-GEP with both MC-GEP and MUC-GEP on optimizing Rastrigin and Griewangk with various dimensions. Experiment results also show that UC-GEP is the best among these three algorithms.

Mining h-dimensional enhanced semantic association rule based on immune-based gene expression programming

Rule mining is very important for data mining. However, traditional association rule is relatively weak in semantic representation. To address it, the main contributions of this paper included: (1) proposing formal concepts on h-Dimensional Enhanced Semantic Association Rule (h-DESAR) with self-contained logic operator; (2) proposing the h-DESAR mining method based on Immune-based Gene Expression Programming (ERIG); (3) presenting some novel key techniques in ERIG. Experimental results showed that ERIG is feasible, effective and stable.

A Novel Personalized Recommendation for Intelligent Sharing of Network Resources

In order to solve the problems such as: producing high quality recommendations, efficient organizing and performing thousands of recommendations per second for millions of users and resources, and achieving high accuracy of recommendation. This paper proposes a novel information feature spatial based personalized recommendation strategy to fulfill intelligent sharing network resources built through network users’ dynamic collection behaviors. The main contributions including: (1) Giving the formula to calculate information rating values of web pages and network users by means of users’ collection behavior; (2) Proposing the construction of information feature spatial based on SHG-Tree to organize, locate and index all network resources in recommendation platform; (3) Proposing four information match algorithms with different index granularity and applying them to six types of personalized recommendation schemes; (4) Applying recommendation methods to a Wushu service network platform, the result shows that the recommendation service can achieve millisecond respond and the recommendation satisfaction can exceed 70%.

An Interest Feature Spatial Approach for Personalized Recommendation

To expand users actions of personalized recommendation, this paper introduces an Interest Feature Spatial based Recommendation Model. This model combines both collection behavior data of network users and content data of web pages located by URL address. The main content includes: (1) Proposing the construction of interest feature spatial based on SHG-Tree; (2) Proposing the formula to calculate interest feature values of network resources; (3) Proposing four interest match algorithms along with six types of personalized recommendation schemes. Experiments show that the recommendation service can achieve millisecond responding, the precision, especially recall metric is better than item-based collaborative filtering algorithm.

Auto-Programming for Numerical Data Based on Remnant-Standard-Deviation-Guided Gene Expression Programming

Automatically numerical data modeling and computer code generation is significant for data mining, data reverse engineering, engineering applications, etc. On auto-programming for numerical data, a new approach, Remnant-standard-Deviation-guided Gene Expression Programming (RD-GEP), was proposed. New individual structure, the K-expression to Reverse Polish Notation code generation without expression tree construction algorithm (K2RPN), and remnant-standard-deviation based fitness evaluation method in RD-GEP were presented and studied. New individual structure makes easy to I/O or storage the candidate solution. New decoding algorithm with linear-time complexity can simplify system operation and unify I/O format. New evaluation mechanism can reduce hypothesis solution space to improve system performance and precision. Feasibility and usability of RD-GEP were verified on various synthetic data sets and real “Fishcatch” data set. Experimental results showed RD-GEP is good at automatically modeling numerical data and generating reverse polish notation for target model.

SHG-Tree: an efficient granularity-based spatial index structure

To improve the access efficiency of multidimensional spatial database, this study proposes a new index structure named Space Hypercube Grid Tree (SHG-Tree). By avoiding the problems of node split and recombination, SHG-Tree can efficiently support the common operations over spatial database containing objects with dynamic region. The main contributions of this paper include: (1) Proposes SHG-Tree of n-dimensional space with a hierarchical tree structure. It reflects the region overlapping relationship of hypercube grid units with different granularity. (2) Proposes the linearization methods to present the bounding rectangle of object as a union of variant granularity hypercube grids. (3) Gives operations of SHG-Tree. Experiments result shows the size of SHG-Tree is small enough to remain in main memory even to very large spatial database by applying proper linearization strategy and the queries on SHG-Tree are less than ten milliseconds to ensure the real-time of query.