Tabular Transformers typically tokenize each feature as a token, so sequence length grows with the number of columns and attention cost scales as O(F2). We introduce EvoPatch, an offline tokenizer that learns a patch tokenization: it groups columns into K patches and represents each patch as a pooled token. EvoPatch builds a leakage-safe feature space from Gauss-rank normalized numeric features and leave-one-out target-encoded categorical features, constructs a feature similarity graph, and searches for a partition with multi-objective evolution. The search maximizes a fast validation proxy while maximizing token reduction, with a memetic refinement step that balances information across patches for stable solutions. The selected partition is computed once per dataset and reused unchanged for both training and inference, reducing attention cost without additional test-time computation and improving the computational efficiency of attention models on wide tables. Used as a tokenizer, EvoPatch compresses inputs for FT-Transformer, TabTransformer, and AutoInt on seven public datasets, achieving 17.5–75.4% token reduction with the largest absolute relative change in accuracy/RMSE bounded by 3.1%, and yields up to 2.8× training and 3.0× inference speedup on the widest dataset.

We propose a novel diffusion-based framework for automatic colorization of Anime-style facial sketches, which preserves the structural fidelity of the input sketch while effectively transferring stylistic attributes from a reference image. Our approach builds upon recent continuous-time diffusion models, but departs from traditional methods that rely on predefined noise schedules, which often fail to maintain perceptual consistency across the generative trajectory. To address this, we introduce SSIMBaD (Sigma Scaling with SSIM-Guided Balanced Diffusion), a sigma-space transformation that ensures linear alignment of perceptual degradation, as measured by structural similarity. This perceptual scaling enforces uniform visual difficulty across timesteps, enabling more balanced and faithful reconstructions. On a large-scale Anime face dataset, SSIMBaD attains state-of-the-art structural fidelity and strong perceptual quality, with robust generalization to diverse styles and structural variations.

Portfolio optimization is a control problem whose objective is to find the optimal strategy for the process of selecting the proportions of assets that can provide the maximum return. Conventional approaches formulate the problem as a single Markov decision process and apply reinforcement learning methods to provide solutions. However, it is well known that financial markets involve non-stationary processes, leading to violations of this assumption in these methods. In this work, we reformulate the portfolio optimization problem to deal with the non-stationary nature of financial markets. In our approach, we divide a long-term process into multiple short-term processes to adapt to context changes and consider the portfolio optimization problem as a multitask control problem. Thereafter, we propose an evolutionary meta reinforcement learning approach to search for an initial policy that can quickly adapt to the upcoming target tasks. We model the policies as convolutional networks that can score the match of the patterns in market data charts. Finally, we test our approach using real-world cryptographic currency data and show that it adapts well to the changes in the market and leads to better profitability.

We suggest a use of genetic programming for transformation from a vector space to an understandable graph representation, which is part of a project to inspect the latent space in matrix factorization. Given a relation matrix, we can apply standard techniques such as non-negative matrix factorization to extract low dimensional latent space in vector representation. While the vector representation of the latent space is useful, it is not intuitive and hard to interpret. The transformation with the help of genetic programming allows us to better understand the underlying latent structure. Applying the method in the context of a stock market, we show that it is possible to recover the tree representation of technical patterns from a relation matrix. Leveraging the properties of the vector representations, we are able to find patterns that correspond to cluster centers of technical patterns. We further investigate the geometry of the latent space.

A neural network-based chart pattern represents adaptive parametric features, including non-linear transformations, and a template that can be applied in the feature space. The search of neural network-based chart patterns has been underexposed despite its potential expressiveness.

Chart pattern analysis uses knowledge extracted from graphical information of price movements. There are two representative types of problems in chart pattern analysis: the matching problem and the search problem. There have been extensive studies on chart pattern matching. However, chart pattern search has not yet drawn much interest. Instead of automatic search, most studies use chart patterns manually designed by financial experts. In this paper, we suggest an automatic algorithm that searches a rule-based chart pattern. We formulate rule-based chart pattern search as an optimization problem for a genetic algorithm. The suggested genetic algorithm includes a considerable amount of problem-specific manipulation. The algorithm successfully found attractive patterns working on the Korean stock market. We studied the rules used in the found patterns, noting that they are rising-support patterns. In addition, the automated pattern generation uses designs at a higher level of abstraction.

Measuring similarity between source codes has lots of applications, such as code plagiarism detection, code clone detection, and malware detection. A variety of methods for the measurement have been developed and program-dependence-graph based methods are known to be well working against disguise techniques. But these methods usually rely on solving NP-hard problems which cause a scalability issue. In this paper, we propose a genetic algorithm to measure the similarity between two codes by solving an error correcting subgraph isomorphism problem on dependence graphs. We propose a new cost function for this problem, which reflects the characteristic of the source codes. An incremental genetic algorithm is used to solve the problem. The size of the graph to be searched is gradually increasing during the evolutionary process. We developed new operators for the algorithm, and the overall system is tested on some real world data. Experimental results showed that the system successfully works on code plagiarism detection and malware detection. The similarity computed by the system turned out to reflect the similarity between the codes properly.

We tackle the problem of knowledge discovery in time series data using genetic programming and GPGPUs. Using genetic programming, various precursor patterns that have certain attractive qualities are evolved to predict the events of interest. Unfortunately, evolving a set of diverse patterns typically takes huge execution time, sometimes longer than one month for this case. In this paper, we address this problem by proposing a parallel GP framework using GPGPUs, particularly in the context of big financial data. By maximally exploiting the structure of the nVidia GPGPU platform on stock market time series data, we were able see more than 250-fold reduction in the running time.

In this paper, we propose a genetic algorithm for a LEGO(R) brick layout problem. The task is to build a given 3D object with LEGO(R) bricks. A brick layout is modeled as a solution to a combinatorial optimization problem, through intermediate voxelization, which tries to maximize the connectivity and then minimize the number of used bricks. We attack the problem in the context of genetic search. The proposed randomized greedy algorithm produces initial solutions, and the solutions are effectively improved by an evolutionary process. New domain-specific methods are proposed as well, which include a random boundary mutation and a thickening approach. We tested our algorithm on various objects collected from the web. Experimental results showed that the algorithm produces efficient, and mostly optimal solutions for benchmark models. Unlike some previous works, our algorithm is not limited to assemble few specific objects, but it can deal with diverse kind of objects. To the best of our knowledge, this is the most extensive empirical study on the problem.

Finding an isomorphic subgraph is a key problem in many real world applications modeled on graph. In this paper, we propose a new hybrid genetic algorithm(GA) for subgraph isomorphism problem which uses an incremental approach. We solve the problem with increasing the size of the subproblem step by step. The graph for which we search is gradually expanded from the empty structure to the entire one. We apply a hybrid GA to each subproblem, initialized with the evolved population of previous step. We present design issues for the incremental approach, and the effects of each design decision are analyzed by experiment. The proposed algorithm is tested on widely used dataset. With apposite vertex reordering along with moderate population diversity, incremental approach brought a significant performance improvement. Experimental results showed that our algorithm outperformed representative previous works.

Meteorological data collected from automatic weather stations have played an important role in forecasting and analyzing a large variety of phenomena. However, abnormal values are abundant in meteorological data due to manifold faults in observation systems. In this paper, we attempt to recover abnormal values. We present three estimation models based on machine learning techniques and compare them with traditional estimation methods, interpolations. Unlike the interpolation methods, which use only the target attribute, the proposed models utilize the additional information consisting of the associated attributes of the target station and the relevant data of the neighbor weather stations. Experiments were conducted for 692 locations in South Korea from 2007 to 2012. The results showed that the proposed approaches estimated target values better than the interpolation methods for all weather elements except one and the additional information helped achieve better performance.

In this paper we propose a multi-objective genetic algorithm for the subgraph isomorphism problem. Usually, the number of different edges between two graphs has been used as a fitness function. This approach has limitations in that it only considers directly-visible characteristics of current solutions, not considering the potential for being an optimal solution. We designed a fitness function in which solutions with higher potential can be rated high. This new fitness function has good properties such as transforming the solution space globally convex and improving the performance of local heuristics and genetic algorithms. Experimental results show that the suggested approach brings a considerable improvement in performance and efficiency.

The graph partitioning problem occurs in numerous applications such as circuit placement, matrix factorization, load balancing, and community detection. For this problem, genetic algorithm is a representative approach with competitive performance with many related papers being published. Although there are a number of surveys on graph partitioning, none of them deals with genetic algorithms in much detail. In this survey, a number of problem-specific issues in applying genetic algorithms to the graph partitioning problem are discussed; the issues include encoding, crossover, normalization, and balancing.

We present a genetic feature selection method for the optimization of an image set for producing real-time photomosaics. A photomosaic is an image that is divided into grid cells, each of which is replaced with another image of an appropriate color pattern. Construction of a photomosaic requires a large set of tile images with different patterns. However, a large number of photo images requires expensive pattern searching and a large storage space. These requirements can cause problems in applications to real-time domains or mobile devices with limited resources. We present a genetic feature selection method for building an optimized image set to accelerate pattern searching and minimize the memory cost.

Computer malware is becoming a serious threat to our daily life in the information-based society. Especially, script malwares has become famous recently, since a wide range of programs supported scripting, the fact that makes such malwares spread easily. Because of viral polymorphism, current malware detection technologies cannot catch up the exponential growth of polymorphic malwares. In this paper, we propose a detection mechanism for script malwares, using dependency graph analysis. Every script malware can be represented by a dependency graph and then the detection can be transformed to the problem finding maximum subgraph isomorphism in that polymorphism still maintains the core of logical structures of malwares. We also present efficient heuristic approaches for maximum subgraph isomorphism, which improve detection accuracy and reduce computational cost. The experimental results of their use in a hybrid GA showed superior detection accuracy against state-of-the-art anti-virus softwares.

The main focus of this paper is to propose integration of dynamic and multiobjective algorithms for graph clustering in dynamic environments under multiple objectives. The primary application is to multiobjective clustering in social networks which change over time. Social networks, typically represented by graphs, contain information about the relations (or interactions) among online materials (or people). A typical social network tends to expand over time, with newly added nodes and edges being incorporated into the existing graph. We reflect these characteristics of social networks based on real-world data, and propose a suitable dynamic multiobjective evolutionary algorithm. Several variants of the algorithm are proposed and compared. Since social networks change continuously, the immigrant schemes effectively used in previous dynamic optimisation give useful ideas for new algorithms. An adaptive integration of multiobjective evolutionary algorithms outperformed other algorithms in dynamic social networks.

We propose a new modular genetic programming for finding attractive and statistically sound technical patterns forstock trading. We restrict the problem space to combinations of modules for more effective space search. We carefully prepared the set of modules based on existing studies of technical indicators and our own experience. Our modular genetic programming successfully found unknown attractive technical patterns for the Korean stock market. A trading simulation with the generated patterns by a commercial tool showed significantly higher accumulative returns than the KOSPI index.

In 1968, Volker Strassen, a young German mathematician, announced a clever algorithm to reduce the asymptotic complexity of$n\times n$matrix multiplication from the order of$n^{3}$to$n^{2.81}$. It soon became one of the most famous scientific discoveries in the 20th century and provoked numerous studies by other mathematicians to improve upon it. Although a number of improvements have been made, Strassen's algorithm is still optimal in his original framework, the bilinear systems of 2$\,\times\,$2 matrix multiplication, and people are still curious how Strassen developed his algorithm. We examined it to see if we could automatically reproduce Strassen's discovery using a search algorithm and find other algorithms of the same quality. In total, we found 608 algorithms that have the same quality as Strassen's, including Strassen's original algorithm. We partitioned the algorithms into nine different groups based on the way they are constructed. This paper was made possible by the combination of genetic search and linear-algebraic techniques. To the best of our knowledge, this is the first work that automatically reproduced Strassen's algorithm, and furthermore, discovered new algorithms with equivalent asymptotic complexity using a search algorithm.

A variant of two-dimensional packing problem was given in the GECCO'2008 competition. This paper describes the genetic algorithm that produced the best result and thus won the No. 1 prize. As the problem is naturally represented by a two-dimensional chromosome, two-dimensional crossovers are used to generate more diverse chromosomes and effectively maintain geographical linkage among genes. We developed a local search heuristic based on the breadth-first search algorithm; we describe how to implement the heuristic efficiently using problem-specific knowledge. The local search was combined with a steady-state genetic algorithm and the combination showed strong synergy.

We propose a genetic algorithm with adaptive elitism-based immigrants which tunes the balance between elitism-based immigrants and random immigrants by itself. Experimental results show that our genetic algorithm with adaptive elitism-based immigrants performs better than that with the elitism-based immigrants for onemax and produces comparable results for royal road and knapsack problems.

We propose a new modular genetic programming for finding attractive and statistically sound technical rules. We restrict the problem space using well-known technical rules to discover attractive technical rules. Experimental results show that our modular genetic programming can successfully find unknown attractive technical rules for Korean stock market.

For a real-valued functionfdefined on {0,1}n, the linkage graph offis a hypergraph that represents the interactions among the input variables with respect tof. In this paper, lower and upper bounds for the number of function evaluations required to discover the linkage graph are rigorously analyzed in the black box scenario. First, a lower bound for discovering linkage graph is presented. To the best of our knowledge, this is the first result on the lower bound for linkage discovery. The investigation on the lower bound is based on Yao's minimax principle. For the upper bounds, a simple randomized algorithm for linkage discovery is analyzed. Based on the Kruskal-Katona theorem, we present an upper bound for discovering the linkage graph. As a corollary, we rigorously prove thatO(n2logn) function evaluations are enough for bounded functions when the number of hyperedges isO(n), which was suggested but not proven in previous works. To see the typical behavior of the algorithm for linkage discovery, three random models of fitness functions are considered. Using probabilistic methods, we prove that the number of function evaluations on the random models is generally smaller than the bound for the arbitrary case. Finally, from the relation between the linkage graph and the Walsh coefficients, it is shown that, for bounded functions, the proposed bounds are eventually the bounds for finding the Walsh coefficients.

The maximum k-coverage problem is a generalized version of covering problems. We introduce the problem formally and analyze its property in relation to the operators of genetic algorithm. Based on the analysis, we propose a new crossover tailored to the maximum k-coverage problem. While traditional n-point crossovers have a problem of requiring repair steps, the proposed crossover has an additional advantage of always producing feasible solutions. We give a comparative analysis of the proposed crossover through experiments.

Normalization transforms one parent genotype to be consistent with the other before crossover. In this paper, we explain how normalization alleviates the difficulties caused by nonsynonymously redundant encodings in genetic algorithms. We define the encodings with maximally nonsynonymous property and prove that the encodings induce uncorrelated search spaces. Extensive experiments for a number of problems show that normalization transforms the uncorrelated search spaces to correlated ones and leads to significant improvement in performance.

The multicampaign assignment problem is a campaign model to overcome the multiple-recommendation problem that occurs when conducting several personalized campaigns simultaneously. In this paper, we propose a Lagrangian method for the problem. The original problem space is transformed to another simpler one by introducing Lagrange multipliers, which relax the constraints of the multicampaign assignment problem. When the Lagrangian vector is supplied, we can compute the optimal solution under this new environment in O(NK2) time, whereNandKare the numbers of customers and campaigns, respectively. This is a linear-time method when the number of campaigns is constant. However, it is not easy to find a Lagrangian vector in exact accord with the given problem constraints. We thus combine the Lagrangian method with a genetic algorithm to find good near-feasible solutions. We verify the effectiveness of our evolutionary Lagrangian approach in both theoretical and experimental viewpoints. The suggested Lagrangian approach is practically attractive for large-scale real-world problems.

Inwook Hwang School of Computer Science & Engineering Seoul National University Sillim-dong, Gwanak-gu Seoul, 151-744, Korea [email protected] Yong-Hyuk Kim Department of Computer Science Kwangwoon University Wolgye-dong, Nowon-gu Seoul, 139-701, Korea [email protected] Byung-Ro Moon School of Computer Science & Engineering Seoul National University Sillim-dong, Gwanak-gu Seoul, 151-744, Korea [email protected]

Keystroke authentication is a biometric method utilizing the typing characteristics of users. In this paper, we propose an evolutionary method for stable keystroke authentication. In the method, typing characteristics of users are represented by n-dimensional vectors and an ellipsoidal hypothesis space, which distinguishes a collection of the timing vectors of a user from those of the others, is evolved by a genetic algorithm. A filtering scheme and an adaptation mechanism are also presented to improve the stability and effectiveness of the proposed method. Empirical results show that the error rates of our method for authentication are reasonably small.

Geometric crossover is a representation-independent generalization of the class of traditional mask-based crossover for binary strings. It is based on the distance of the search space seen as a metric space. Although real-code representation allows for a very familiar notion of distance, namely the Euclidean distance, there are also other distances suiting it. Also, topological transformations of the real space give rise to further notions of distance. In this paper, we study the geometric crossovers associated with these distances in a formal and very general setting and show that many preexisting genetic operators for the real-code representation are geometric crossovers. We also propose a novel methodology to remove the inherent bias of pre-existing geometric operators by formally transforming topologies to have the same effect as gluing boundaries.

Geometric crossover is a representation-independent generalization of the traditional crossover defined using the distance of the solution space. By choosing a distance firmly rooted in the syntax of the solution representation as a basis for geometric crossover, one can design new crossovers for any representation. Using a distance tailored to the problem at hand, the formal definition of geometric crossover allows us to design new problem-specific crossovers that embed problem-knowledge in the search. The standard encoding for multiway graph partitioning is highly redundant: each solution has a number of representations, one for each way of labeling the represented partition. Traditional crossover does not perform well on redundant encodings. We propose a new geometric crossover for graph partitioning based on a labeling-independent distance that filters out the redundancy of the encoding. A correlation analysis of the fitness landscape based on this distance shows that it is well suited to graph partitioning. A second difficulty with designing a crossover for multiway graph partitioning is that of feasibility: in general recombining feasible partitions does not lead to feasible offspring partitions. We design a new geometric crossover for permutations with repetitions that naturally suits partition problems and we test it on the graph partitioning problem. We then combine it with the labeling-independent crossover and obtain a much superior geometric crossover inheriting both advantages.

In optimization problems, the contribution of a variable to fitness often depends on the states of other variables. This phenomenon is referred to as epistasis or linkage. In this paper, we show that a new theory of epistasis can be established on the basis of Shannon's information theory. From this, we derive a new epistasis measure called entropic epistasis and some theoretical results. We also provide experimental results verifying the measure and showing how it can be used for designing efficient evolutionary algorithms.

The properties of symmetric fitness functions are investigated. We show that the search spaces obtained from symmetric functions have the zero-correlation structures between fitness and distance. It is also proven that symmetric functions induce a class of the hardest problems in terms of the epistasis variance and its variants. These analyses suggest that the existing quantitative measures cannot discriminate among symmetric functions, which reveals critical limitations of the measures. To take a closer look at the symmetric functions, additional analyses are performed from other viewpoints including additive separability and boundedness. It is shown that additive separability in a symmetric function is closely related to the symmetry of its subfunctions. This elucidates why most of the well-known symmetric fitness functions are additively inseparable. The properties of two-bounded symmetric functions are investigated and they are utilized in designing an efficient algorithm to check additive separability for the two-bounded functions. Throughout this paper, we heavily use the generalized Walsh transform over multary alphabets. Our results have an independent interest as a nontrivial application of the generalized Walsh analysis.

In this paper, we propose a hybrid neurogenetic system for stock trading. A recurrent neural network (NN) having one hidden layer is used for the prediction model. The input features are generated from a number of technical indicators being used by financial experts. The genetic algorithm (GA) optimizes the NN's weights under a 2-D encoding and crossover. We devised a context-based ensemble method of NNs which dynamically changes on the basis of the test day's context. To reduce the time in processing mass data, we parallelized the GA on a Linux cluster system using message passing interface. We tested the proposed method with 36 companies in NYSE and NASDAQ for 13 years from 1992 to 2004. The neurogenetic hybrid showed notable improvement on the average over the buy-and-hold strategy and the context-based ensemble further improved the results. We also observed that some companies were more predictable than others, which implies that the proposed neurogenetic hybrid can be used for financial portfolio construction.

The properties of symmetric fitness functions are investigated. We show that a well-known encoding scheme inducing symmetric functions has the non-synonymous property and the search spaces obtained from symmetric functions have the zero-correlation structures. The Walsh analysis reveals the properties of symmetric functions related to additive separability, problem difficulty measures and so on. Our results support the claim of other researchers that the search spaces with symmetry induce relatively difficult problems. The results also present some limitations of existing problem difficulty measures for symmetric fitness functions.

We propose a new gene reordering scheme for the graph bisection problem. Our gene reordering starts with two or more vertices to capture the clustering structure of graphs effectively. We devised a chromosome repairing method for hybrid genetic search, which helps exploit clusters when combined with gene reordering. Experimental tests showed that the suggested reordering scheme significantly improves the performance of genetic algorithms over previous reordering methods.

Geometric crossover is a representation-independent generalization of the traditional crossover defined using the distance of the solution space. Using a distance tailored to the problem at hand, the formal definition of geometric crossover allows to design new problem-specific crossovers that embed problem-knowledge in the search. The standard encoding for multiway graph partitioning is highly redundant: each solution has a number of representations, one for each way of labeling the represented partition. Traditional crossover does not perform well on redundant encodings. We propose a new geometric crossover for graph partitioning based on a labeling-independent distance that filters the redundancy of the encoding. A correlation analysis of the fitness landscape based on this distance shows that it is well suited to graph partitioning. Our new genetic algorithm outperforms existing ones.

Function approximation or regression is the problem of finding a function that best explains the relationship between independent variables and a dependent variable from the observed data. Genetic programming has been considered a promising approach for the problem since it is possible to optimize both the functional form and the coefficients. Genetic programming has been considered a promising approach for function approximation since it is possible to optimize both the functional form and the coefficients. However, it is not easy to find an optimal set of coefficients by using only non-adjustable constant nodes in genetic programming. To overcome the problem, there have been some studies on genetic programming using adjustable parameters in linear or nonlinear models. Although the nonlinear parametric model has a merit over the linear parametric model, there have been few studies on it. In this paper, we propose a nonlinear parametric genetic programming which uses a nonlinear gradient method to estimate parameters. The most notable feature in the proposed genetic programming is that we design a parameter attachment algorithm using as few redundant parameters as possible. It showed a significant performance improvement over the traditional genetic programming approaches on real-world application problems.

This paper proposes a hybrid genetic algorithm for multiple sequence alignment. The algorithm evolves guide sequences and aligns input sequences based on the guide sequences. It also embeds a local search heuristic to search the problem space effectively. In the experiments for various data sets, the proposed algorithm showed the performance comparable to existing algorithms.

We propose a new crossover that generalizes cycle crossover to permutations with repetitions and naturally suits partition problems. We tested it on graph partitioning problems obtaining excellent results.

It is crucial to maximize targeting efficiency and customer satisfaction in personalized marketing. State-of-the-art techniques for targeting focus on the optimization of individual campaigns. Our motivation is the belief that the effectiveness of a campaign with respect to a customer is affected by how many precedent campaigns have been recently delivered to the customer. We raise the multiple recommendation problem, which occurs when performing several personalized campaigns simultaneously. We formulate the multicampaign assignment problem to solve this issue and propose algorithms for the problem. The algorithms include dynamic programming and efficient heuristic methods. We verify by experiments the effectiveness of the problem formulation and the proposed algorithms.

We propose a genetic fuzzy discretization method for continuous numerical attributes. Traditional discretization methods categorize the continuous attributes into a number of bins. Because they are made on crisp discretization, there exists considerable information loss. Fuzzy discretization allows overlapping intervals and reflects linguistic classification. However, the number of intervals, the boundaries of intervals, and the degrees of overlapping are intractable to get optimized. We use a genetic algorithm to optimize these parameters. Experimental results showed considerable improvement on the classification accuracy over a crisp discretization and a typical fuzzy discretization.

Jung-Hwan Kim School of Computer Science and Engineering Seoul National University Shilim-dong, Kwanak-gu, Seoul, 151-742 Korea [email protected] Sung-Soon Choi School of Computer Science and Engineering Seoul National University Shilim-dong, Kwanak-gu, Seoul, 151-742 Korea [email protected] Byung-Ro Moon School of Computer Science and Engineering Seoul National University Shilim-dong, Kwanak-gu, Seoul, 151-742 Korea [email protected]

We propose three distance measures for genetic search space. One is a distance measure in the population space that is useful for understanding the working mechanism of genetic algorithms. Another is a distance measure in the solution space for K-grouping problems. This can be used for normalization in crossover. The third is a level distance measure for genetic algorithms, which is useful for measuring problem difficulty with respect to genetic algorithms. We show that the proposed measures are metrics and the measures are efficiently computed.

In this paper, we propose a neuro-genetic stock prediction system based on financial correlation between companies. A number of input variables are produced from the relatively highly correlated companies. The genetic algorithm selects a set of informative input features among them for a recurrent neural network. It showed notable improvement over not only the buy-and-hold strategy but also the recurrent neural network using only the input variables from the target company.

Feature extraction is a process that extracts salient features from observed variables. It is considered a promising alternative to overcome the problems of weight and structure optimization in artificial neural networks. There were many nonlinear feature extraction methods using neural networks but they still have the same difficulties arisen from the fixed network topology. In this paper, we propose a novel combination of genetic algorithm and feedforward neural networks for nonlinear feature extraction. The genetic algorithm evolves the feature space by utilizing characteristics of hidden neurons. It improved remarkably the performance of neural networks on a number of real world regression and classification problems.

The interaction among variables of an optimization problem is known as epistasis, and its degree is an important measure for the nonlinearity of the problem. We address the problem of enormous time complexity for computing Davidor's epistasis variance of the traveling salesman problem (TSP). To reduce the complexity, we introduce the concept of schema-linear problem (SLP), show that TSP is a SLP, and present a relevant lemma, called Summation Rule. Using the Summation Rule, we provide a closed formula for epistasis that reduces the time complexity from O(nn) to O(n2). Additionally, we propose a new more scalable measure of epistasis by a careful derivation from the original.

We propose a new evolutionary approach to solve the 0/1 multiple knapsack problem. We approach the problem from a new viewpoint different from traditional methods. The most remarkable feature is the Lagrangian method. Lagrange multipliers transform the problem, keeping the optimality as well as decreasing the complexity. However, it is not easy to find Lagrange multipliers nearest to the constraints of the problem. We propose an evolution strategy to find the optimal Lagrange multipliers. Also, we improve the evolution strategy by adjusting its objective function properly. We show the efficiency of the proposed methods by the experiments. We make comparisons with existing general approach on well-known benchmark data.

A hybrid genetic algorithm (GA) is proposed for the optimal sorting network problem. Based on a graph-theoretical viewpoint, we devised a solution repair heuristic which incorporates a strong local optimization. We also propose a new encoding scheme which combines the characteristics of Lamarckian and Baldwinian GAs. Using a single-CPU PC, we obtained results comparable to previous results obtained with supercomputers.

This paper proposes a novel hybrid genetic algorithm for feature selection. Local search operations are devised and embedded in hybrid GAs to fine-tune the search. The operations are parameterized in terms of their fine-tuning power, and their effectiveness and timing requirements are analyzed and compared. The hybridization technique produces two desirable effects: a significant improvement in the final performance and the acquisition of subset-size control. The hybrid GAs showed better convergence properties compared to the classical GAs. A method of performing rigorous timing analysis was developed, in order to compare the timing requirement of the conventional and the proposed algorithms. Experiments performed with various standard data sets revealed that the proposed hybrid GA is superior to both a simple GA and sequential search algorithms.

In the field of evolutionary algorithms (EAs), many operators have been introduced for the traveling salesman problem (TSP). Most encoding schemes have various restrictions that often result in a loss of information contained in problem instances. We suggest a new chromosomal encoding scheme that pursues minimal information loss and a crossover scheme with minimal restriction for the two-dimensional (2D) Euclidean TSP. The most notable feature of the suggested crossover is that it uses the 2D tour images themselves for chromosomal cutting. We prove the theoretical validity of the new crossover by an equivalence-class analysis. The proposed encoding/crossover pair outperformed both distance-preserving crossover and edge-assembly crossover, which represent two state-of-the-art crossovers in the literature. We also tested its performance on a mixed framework that incorporates a large-step Markov chain technique into the framework of a traditional EA.

Typical evolutionary algorithms (EAs) exploit the different space-search properties of variation operators, such as crossover, mutation and local optimization. There are also various operators in each element. This paper provides an extensive empirical study on the synergy among multiple crossover operators. We choose a number of different crossover operators in an EA and investigate whether or not their combinations outperform the sole usage of the best crossover operator. The traveling salesman problem and the graph bisection problem were chosen for experimentation. Strong synergy effects were observed in both problems.

We consider the problem space of a sorting network as a graph. Based on this viewpoint, we devised local optimization heuristics for sorting networks and combined them with a genetic algorithm framework. Using a single-CPU PC, we obtained results comparable with previous results obtained using supercomputers.

A genetic algorithm for partitioning a hypergraph into two disjoint graphs of minimum ratio cut is presented. As the Fiduccia-Mattheyses graph partitioning heuristic turns out to be not effective when used in the context of a hybrid genetic algorithm, we propose a modification of the Fiduccia-Mattheyses heuristic for more effective and faster space search by introducing a number of novel features. We also provide a preprocessing heuristic for genetic encoding designed solely for hypergraphs which helps genetic algorithms exploit clustering information of input graphs. Supporting combinatorial arguments for the new preprocessing heuristic are also provided. Experimental results on industrial benchmarks circuits showed visible improvement over recently published algorithms with a lower growth rate of running time.

This paper suggests a new genetic algorithm (GA) for VLSI circuit partitioning problem. In a genetic algorithm, the encoding of a solution plays an important role. The key feature of the new genetic algorithm is a technique to provide dynamically many encodings in which encodings themselves undergo evolution. Before generating every new solution, we first generate a new encoding by combining two encodings chosen from a pool containing diverse encodings. The new solution is generated by a crossover which combines two parent solutions which are temporarily encoded by the generated encoding scheme. That is, a new solution is generated by a two-layered crossover. Depending on the new solution's quality and its improvement over the parents solutions, a fitness value is assigned to the underlying encoding. The encoding is discarded or enter the pool based on the fitness. Two populations are maintained for this purpose: one for solutions and the other for diverse encodings. On experiments with the public ACM/SIGDA benchmark circuits, the new genetic algorithm significantly outperformed recently published state-of-the-art approaches.

Hybrid genetic algorithms (GAs) for the graph partitioning problem are described. The algorithms include a fast local improvement heuristic. One of the novel features of these algorithms is the schema preprocessing phase that improves GAs' space searching capability, which in turn improves the performance of GAs. Experimental tests on graph problems with published solutions showed that the new genetic algorithms performed comparable to or better than the multistart Kernighan-Lin algorithm and the simulated annealing algorithm. Analyses of some special classes of graphs are also provided showing the usefulness of schema preprocessing and supporting the experimental results.

Article A fast and stable hybrid genetic algorithm for the ratio-cut partitioning problem on hypergraphs Share on Authors: Thang Nguyen Bui Dept. of Computer Science, Pennsylvania State University, Middletown, PA Dept. of Computer Science, Pennsylvania State University, Middletown, PAView Profile , Byung Ro Moon Dept. of Computer Science and Engineering, Pennsylvania State University, University Park, PA Dept. of Computer Science and Engineering, Pennsylvania State University, University Park, PAView Profile Authors Info & Claims DAC '94: Proceedings of the 31st annual Design Automation ConferenceJune 1994 Pages 664–669https://doi.org/10.1145/196244.196607Published:06 June 1994 20citation407DownloadsMetricsTotal Citations20Total Downloads407Last 12 Months0Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access