Graph Neural Networks (GNNs) have emerged as powerful tools for learning on graph-structured data, but often struggle to balance local and global information. While graph Transformers aim to address this by enabling long-range interactions, they often overlook the inherent locality and efficiency of Message Passing Neural Networks (MPNNs). We propose a new concept called 'fractal nodes', inspired by the fractal structure observed in real-world networks. Our approach is based on the intuition that graph partitioning naturally induces fractal structure, where subgraphs often reflect the connectivity patterns of the full graph. Fractal nodes are designed to coexist with the original nodes and adaptively aggregate subgraph-level feature representations, thereby enforcing feature similarity within each subgraph. We show that fractal nodes alleviate the over-squashing problem by providing direct shortcut connections that enable long-range propagation of subgraph-level representations. Experiment results show that our method improves the expressive power of MPNNs and achieves comparable or better performance to graph Transformers while maintaining the computational efficiency of MPNN by improving the long-range dependencies of MPNN.

Continual learning methods used to force neural networks to process sequential tasks in isolation, preventing them from leveraging useful inter-task relationships and causing them to repeatedly relearn similar features or overly differentiate them. To address this problem, we propose a fully differentiable, exemplar-free expandable method composed of two complementary memories: One learns common features that can be used across all tasks, and the other combines the shared features to learn discriminative characteristics unique to each sample. Both memories are differentiable so that the network can autonomously learn latent representations for each sample. For each task, the memory adjustment module adaptively prunes critical slots and minimally expands capacity to accommodate new concepts, and orthogonal regularization enforces geometric separation between preserved and newly learned memory components to prevent interference. Experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet show that the proposed method outperforms 14 state-of-the-art methods for class-incremental learning, achieving final accuracies of 55.13%, 37.24%, and 30.11%, respectively. Additional analysis confirms that, through effective integration and utilization of knowledge, the proposed method can increase average performance across sequential tasks, and it produces feature extraction results closest to the upper bound, thus establishing a new milestone in continual learning.

Consistent reasoning about 3D spatial relations across changing viewpoints is fundamental for Embodied AI agents operating in dynamic environments.While Large Vision-Language Models (LVLMs) have advanced multimodal perception, their ability to maintain spatial consistency across diverse perspectives remains underexplored.Existing benchmarks primarily assess spatial capabilities from a static, single-view, and egocentric perspective, failing to capture the dynamic nature of real-world spatial cognition.To address this gap, we introduce SCOPE (Spatial COnsistency across PErspectives and Viewpoints), a comprehensive benchmark designed to rigorously diagnose spatial reasoning capabilities.Grounded in human cognitive theories of dual spatial representations, SCOPE discretizes the 360 • field into multiview scenarios to systematically evaluate both allocentric and egocentric reasoning capabilities.Our dataset comprises 20.1K spatial VQA pairs derived from high-quality 3D environments.Through an extensive evaluation of 26 state-of-the-art LVLMs, we identify two fundamental limitations that prevent consistent spatial understanding across viewpoints.We hope SCOPE facilitates the diagnosis of spatial reasoning, serving as a stepping stone toward reliable embodied action.

Recent advances in large language models (LLMs) have improved logical reasoning by incorporating formal logic or explicit structured representations.However, such methods often lose track of what is true now in multistep reasoning, failing to maintain a coherent global state and its logical consequences.Motivated by Situation Model Theory in cognitive psychology, which views comprehension as constructing and updating a mental model of events along key dimensions (time, space, causality, intention, protagonist), we propose a cognitively inspired method of Situation Working Memory (SituW) for contextual reasoning in LLMs.SituW first builds a situation representation by decomposing text along these five dimensions, and guides LLM inference with the evolving state.Keeping an explicit, dynamically updated situation memory instead of a static logical form encourages globally consistent reasoning over the situation model rather than raw text.Evaluated in both supervised and prompt-based settings, SituW improves accuracy by 23.3%p and 15.93%p while reducing "uncertain" predictions, suggesting that explicit situation modeling supports more globally consistent LLM reasoning.Our code is available at

Rapid urbanization and population growth raise significant challenges in modern traffic management, where traffic prediction is essential for intelligent transportation systems . Recent proliferation of graph neural networks also could not give us the satisfactory solution, because predicting traffic flow requires effective modeling of complex spatial correlations and temporal dependencies among sensors. In this paper, we propose a novel graph transformer that mitigates the spatial and temporal heterogeneity simultaneously. Graph partitioning to capture spatial heterogeneity induces the nodes grouped with similar contextual properties. The proposed transformer effectively handles long-term temporal dependencies, and combines subgraph embeddings to represent the correlation of global patterns. Experimental results on four traffic prediction benchmark datasets with high spatial dependencies show that the proposed method produces a 12.33%p performance improvement against the 14 state-of-the-art methods. Especially, it exhibits excellent performance in 60-minute predictions, and training times are comparable to the competitive methods.

Deep neural networks struggle with incremental updates due to catastrophic forgetting, where newly acquired knowledge interferes with the learned previously. Continual learning (CL) methods aim to overcome this limitation by effectively updating the model without losing previous knowledge, but they find it difficult to continuously maintain knowledge about previous tasks, resulting from overlapping stored information. In this paper, we propose a CL method that preserves previous knowledge as multivariate Gaussian distributions by independently storing the model's outputs per class and continually reproducing them for future tasks. We enhance the discriminability between classes and ensure the plasticity for future tasks by exploiting contrastive learning and representation regularization. The class-wise spatial means and covariances, distinguished in the latent space, are stored in memory, where the previous knowledge is effectively preserved and reproduced for incremental tasks. Extensive experiments on benchmark datasets such as CIFAR-10, CIFAR-100, and ImageNet-100 demonstrate that the proposed method achieves accuracies of 93.21%, 77.57%, and 78.15%, respectively, outperforming state-of-the-art CL methods by 2.34 %p, 2.1 %p, and 1.91 %p. Additionally, it achieves the lowest mean forgetting rates across all datasets.

3D sandbox games like Minecraft allow users to perform innumerable actions in diverse and complex environments. The design of autonomous agents in various environments is crucial for real-world application and utilization. Generative adversarial imitation learning (GAIL) is designed to replicate human demonstrations in more complex and subtle environments and solve various tasks by autonomous agents, but it has limitations in processing complex sequence input and sufficiently suppressing incorrect policies, making it insufficient for generating complex actions for complex inputs in Minecraft. In this paper, we propose an extended GAIL that effectively addresses Minecraft tasks, considering the complexity of the inputs and outputs. The proposed method, using a global encoder shared between the agent and discriminator, ensures robustness for image sequences and improves the positive reward function to include negative values, enabling the agent to construct optimal trajectories among a variety of actions. Experiments on Minecraft game scenarios confirm the superiority of the proposed method over typical GAILs, achieving human-like scores in Navigate and TreeChop tasks. Additionally, we reveal that it is performed in a typical benchmark environment to show that it can operate in game domains. Ablation studies demonstrate that (1) the global encoder can extract meaningful representations from raw image sequence inputs, (2) the reward function significantly impacts the performance in composing actions available to the autonomous agent, and (3) the proposed method, through its adjustment of input and output in GAIL, can be applied across various GAIL-based models.

Recent research in the field of graph neural network (GNN) has identified a critical issue known as "over-squashing," resulting from the bottleneck phenomenon in graph structures, which impedes the propagation of long-range information. Prior works have proposed a variety of graph rewiring concepts that aim at optimizing the spatial or spectral properties of graphs to promote the signal propagation. However, such approaches inevitably deteriorate the original graph topology, which may lead to a distortion of information flow. To address this, we introduce an ex**pand**ed width-**a**ware (**PANDA**) message passing, a new message passing paradigm where nodes with high centrality, a potential source of over-squashing, are selectively expanded in width to encapsulate the growing influx of signals from distant nodes. Experimental results show that our method outperforms existing rewiring methods, suggesting that selectively expanding the hidden state of nodes can be a compelling alternative to graph rewiring for addressing the over-squashing.

Anomaly detection and state prediction research using multivariate data is being actively conducted in various industrial fields. However, since most dynamically operating industrial machines perform different operating conditions, they contain different types of abnormal conditions, making it difficult to detect anomalies and predict the remaining life. This white paper proposes a condition diagnosis model based on multi-sensor data prediction for estimating the remaining lifespan of equipment while solving two complex problems of detecting four typical abnormal conditions and sensor omissions of industrial machines. First, we use a multi-sensor data generation model to learn relationships between sensors, and second, we use a sensor data prediction model to learn sensor-specific feature information. In order to extract the temporal and spatial characteristics of sensing information and to derive the relationship between the sensors, we propose an attention model with three types of cases. Finally, the state of the device is diagnosed through the difference between the model predicted value and the actual value, and future state information of the device is predicted through the accumulation of error information. In order to prove the robustness of the proposed model, extensive experiments were conducted focusing on the case where sensor omission occurred due to data from equipment with more than 4 types and conditions. Our model produces missing sensor data with about 92% accuracy and detects anomalies with about 88% accuracy, even if parts of the sensor are missing or the operating environments have been changed. The proposed model has improved anomaly detection accuracy compared to the comparative model, and has been proven to be applicable to real industrial problems.

Abusive language detection models tend to have a gender bias problem in which the model is biased towards sentences containing identity words of specific gender groups. Previous studies to reduce bias, such as projection methods, tend to lose information in word vectors and sentence context, resulting in low detection accuracy. This paper proposes a novel method that mitigates gender bias while preserving original information by regularizing sentence embedding vectors based on information theory. Latent vectors generated by an autoencoder are debiased through dual regularization using a gender discriminator, an abuse classifier, and a decoder. While the gender discriminator labels are randomized, the discriminator confuses the gender feature, and the classifier retains the abuse information. Latent vectors are regularized through information theoretic adversarial optimization that disentangles and mitigates gender features. We show that the proposed method successfully orthogonalizes the direction of the correlated information and reduces the gender feature through calculation of subspaces and embedding vector visualization. Moreover, the proposed method maintains the highest accuracy among the four state-of-the-art bias mitigation methods and shows superior performance in reducing gender bias in four different Twitter datasets for abusive language detection.

Graph neural networks (GNNs) are one of the most popular research topics for deep learning. GNN methods typically have been designed on top of the graph signal processing theory. In particular, diffusion equations have been widely used for designing the core processing layer of GNNs, and therefore they are inevitably vulnerable to the notorious oversmoothing problem. Recently, a couple of papers paid attention to reaction equations in conjunctions with diffusion equations. However, they all consider limited forms of reaction equations. To this end, we present a reaction-diffusion equation-based GNN method that considers all popular types of reaction equations in addition to one special reaction equation designed by us. To our knowledge, our paper is one of the most comprehensive studies on reaction-diffusion equation-based GNNs. In our experiments with 9 datasets and 28 baselines, our method, called GREAD, outperforms them in a majority of cases. Further synthetic data experiments show that it mitigates the oversmoothing problem and works well for various homophily rates.

Collaborative filtering is one of the most fundamental topics for recommender systems. Various methods have been proposed for collaborative filtering, ranging from matrix factorization to graph convolutional methods. Being inspired by recent successes of graph filtering-based methods and score-based generative models (SGMs), we present a novel concept of blurring-sharpening process model (BSPM). SGMs and BSPMs share the same processing philosophy that new information can be discovered (e.g., new images are generated in the case of SGMs) while original information is first perturbed and then recovered to its original form. However, SGMs and our BSPMs deal with different types of information, and their optimal perturbation and recovery processes have fundamental discrepancies. Therefore, our BSPMs have different forms from SGMs. In addition, our concept not only theoretically subsumes many existing collaborative filtering models but also outperforms them in terms of Recall and NDCG in the three benchmark datasets, Gowalla, Yelp2018, and Amazon-book. In addition, the processing time of our method is comparable to other fast baselines. Our proposed concept has much potential in the future to be enhanced by designing better blurring (i.e., perturbation) and sharpening (i.e., recovery) processes than what we use in this paper. Our code is available at https://github.com/jeongwhanchoi/BSPM.

Recently, deep learning models proliferate in the prediction of power demand for efficient planning of power consumption. However, the “black-box” characteristics of deep learning hinders from establishing a specific plan because it cannot explain the cause of the prediction. Recently, there are several attempts to explain the result of deep learning through the analysis of the input attributes that influence the prediction, but they lack of appropriate explanation because of ignoring the time-series property of the input data. In this paper, we propose a deep learning model to explain the impact of the input attributes on the prediction by taking account of the long-term and short-term properties of the time-series forecasting. The model consists of (i) two encoders to represent the power information for prediction and explanation, (ii) a decoder to predict the power demand from the concatenated outputs of encoders, and (iii) an explainer to identify the most significant attributes for predicting the energy consumption. Kullback–Leibler divergence in the loss function induces the long-term and short-term dependencies in latent space constructed by the second encoder. Several experiments on the benchmark dataset of household electric energy demand show that the proposed method explains the prediction appropriately with the most influential input attributes in the long-term and short-term dependencies. We can trade off the gain of the time-series explanation of the result against a slight degradation of the prediction performance.

Recommender systems are a long-standing research problem in data mining and machine learning. They are incremental in nature, as new user-item interaction logs arrive. In real-world applications, we need to periodically train a collaborative filtering algorithm to extract user/item embedding vectors and therefore, a time-series of embedding vectors can be naturally defined. We present a time-series forecasting-based upgrade kit (TimeKit), which works in the following way: it i) first decides a base collaborative filtering algorithm, ii) extracts user/item embedding vectors with the base algorithm from user-item interaction logs incrementally, e.g., every month, iii) trains our time-series forecasting model with the extracted time-series of embedding vectors, and then iv) forecasts the future embedding vectors and recommend with their dot-product scores owing to a recent breakthrough in processing complicated time-series data, i.e., neural controlled differential equations (NCDEs). Our experiments with four real-world benchmark datasets show that the proposed time-series forecasting-based upgrade kit can significantly enhance existing popular collaborative filtering algorithms.

Fall detection is a very challenging task that has a clear impact in the autonomous living of the elderly individuals: suffering a fall with no support increases the fears of the elderly population to continue living by themselves. This study proposes the use of a non-invasive tri-axial accelerometer device placed on a wrist to measure the movements of the participant. The novelty of this study is two fold: on the one hand, the use of a Long-Short Term Memory Neural Network (LSTM) for classification of the Time Series and, on the other hand, the proposal of a novel data augmentation stage that introduces variability in the training by merging the Time Series gathered from both human activities of daily living. The experimentation shows that the combination of a LSTM model together with the data augmentation produces more robust and accurate models that perfectly cope with the validation stage; the high impact fall event detection can be considered solved.

Considering the fatality of phishing attacks that are emphasized by many organizations, the inductive learning approach using reported malicious URLs has been verified in the field of deep learning. However, the deep learning-based method mainly focused on the fitting of a classification task via historical URL observation shows a limitation of recall due to the characteristics of zero-day attack. In order to model the nature of a zero-day phishing attack in which URL addresses are generated and discarded immediately, an approach that utilizes the expert knowledge is promising. We introduce the integration method of deep learning and logic programmed domain knowledge to inject the real-world constraints. We design neural and logic classifiers and propose the joint learning method of each component based on the traditional neuro-symbolic integration. Extensive experiments on three real-world datasets consisting of 222,541 URLs showed the highest recall among the latest deep learning methods, despite the hostile class-imbalanced condition. We demonstrate that the optimized weighting between neural and logic component has an effect of improving the recall over 3% compared to the existing methods.

The classification task of impact noise on vehicle steering system mainly addresses the issue of modeling the transient and impulsive nature. Though various deep learning models including triplet network have been developed, the existing triplet network based on Euclidean distance metric is limited due to the simplicity of distance measure against reverberation generated from the narrow interior space and the low frequency difference generated from the interior finishes. In this paper, we propose a method to overcome the above two major hurdles by modify a sampling algorithm of triplet pairs based on structural similarity index instead of naive Euclidean distance within Monte Carlo based sampling strategy. We verify the proposed modified triplet loss through cross-validation that the proposed sampling method has more than 3% of accuracy improvement with computational cost reduction against the existing triplet networks. The detailed analysis shows that the proposed method can potentially compensate for the disjoint issues between the learning and validation vehicle types.

As the surveillance devices proliferate, various machine learning approaches for video anomaly detection have been attempted. We propose a hybrid deep learning model composed of a video feature extractor trained by generative adversarial network with deficient anomaly data and an anomaly detector boosted by transferring the extractor. Experiments with UCSD pedestrian dataset show that it achieves 94.4% recall and 86.4% precision, which is the competitive performance in video anomaly detection.

Considering the trend of the vehicle market where the vehicle becomes quieter, in-vehicle rattling noise is significant criterion for the quality of the vehicle. Though the latest deep learning algorithms have been introduced for classifying in-vehicle rattling noise, there are limitations due to impulsive and transient nature of rattling noise and reflective and refractive characteristics of in-vehicle environment. In this paper, we propose a novel beamforming method that extracts intra-interchannel spatial features by parameterizing the optimal beamforming weights including Direction-of-Arrival (DOA) function to overcome the addressed problem. The proposed method outperformed the existing deep learning algorithms with 0.9270 accuracy and verified by 10-fold cross validation and chi-squared test. In addition, it is shown that the time cost for classification of rattling noise is appropriate for real-time classification as a side-effect of using convolution-pooling operations.

Point-of-interest (POI) recommendation can help providing better user experience, and provide users with third-party information about restaurant or entertainment. There are several studies to predict the next POI where the user will go so as to recommend appropriate services. They use additional information such as text or location for more precise prediction, or manually define user patterns. However, it is costly to collect and analyze large amounts of data for POI recommendation. In this paper, we propose a novel method to recommend POI by extracting the personalized movement pattern only from the POI data without any additional information. We collected POI data ofl. 5M users for six months from smart card, and produce personalized POI and user embedding. Since it is hard to construct one POI recommendation model for 1.5 million people, we divide them to several groups according to their simple mobility pattern. Given a previous POI sequence, user and group id, the proposed model is trained to maximize the probability of the next POI. Although the learning method of the proposed model is simple, even if the given POI sequence is the same, successive POI can be predicted differently according to the user, resulting in personalized POI recommendation. The proposed model achieves 73.64%, 88.65%, and 91.54% in top-1, 3 and 5 accuracies which are higher than the performance of the baseline model (59.48%, 75.85%, and 80.1%, respectively). Besides, we verify the embedding performance of the proposed model through arithmetic operations between POI vectors.

In this paper, we propose an algorithm to generate optimal trajectory path considering the complex operating environment of excavator front part composed of the boom, arm, and bucket by using genetic algorithm. In order to express motion in space, we propose a method of coordinate plane space of grid cell, and define the fitness value by path distance. After generating chromosome candidates for each motion unit based on the polygonal structure of the front part of the excavator, we calculate the fitness value about each chromosome. The crossover and mutation operations between the chromosomes selected through roulette wheel of top 20% are repeatedly performed to generate paths with optimal fitness values. This paper verifies the structural analysis of the front part of excavator and the utility of the genetic algorithm to optimize the path in the grid space.

Recently, deep learning is one of the most popular techniques in artificial intelligence. However, to construct a deep learning model, various components must be set up, including activation functions, optimization methods, a configuration of model structure called hyperparameters. As they affect the performance of deep learning, researchers are working hard to find optimal hyperparameters when solving problems with deep learning. Activation function and optimization technique play a crucial role in the forward and backward processes of model learning, but they are set up in a heuristic way. The previous studies have been conducted to optimize either activation function or optimization technique, while the relationship between them is neglected to search them at the same time. In this paper, we propose a novel method based on genetic programming to simultaneously find the optimal activation functions and optimization techniques. In genetic programming, each individual is composed of two chromosomes, one for the activation function and the other for the optimization technique. To calculate the fitness of one individual, we construct a neural network with the activation function and optimization technique that the individual represents. The deep learning model found through our method has 82.59% and 53.04% of accuracies for the CIFAR-10 and CIFAR-100 datasets, which outperforms the conventional methods. Moreover, we analyze the activation function found and confirm the usefulness of the proposed method.

Recently, there have been many attempts to predict residential energy consumption using artificial neural networks. The optimization of these neural networks depends on the trial and error of the operator that lacks prior knowledge. They are also influenced by the initial values of the model based on the gradient algorithm and the size of the search space. In this paper, different kinds of hyperparameters are automatically determined by integrating particle swarm optimization (PSO) to CNN-LSTM network for forecasting energy consumption. Our findings reveal that the proposed optimization strategy can be used as a promising alternative prediction method for high prediction accuracy and better generalization capability. PSO achieves effective global exploration by eliminating crossover and mutation operations compared to genetic algorithms. To verify the usefulness of the proposed method, we use the household power consumption data in the UCI repository. The proposed PSO-based CNN-LSTM method explores the optimal prediction structure and achieves nearly perfect prediction performance for energy prediction. It also achieves the lowest mean square error (MSE) compared to conventional machine learning methods.

Abstract In peer‐to‐peer lending, it is important to predict the repayment of the borrower to reduce the lender's financial loss. However, it is difficult to design a powerful feature extractor for predicting the repayment as user and transaction data continue to increase. Convolutional neural networks automatically extract useful features from big data, but they use only high‐level features; hence, it is difficult to capture a variety of representations. In this study, we propose a deep dense convolutional network for repayment prediction in social lending, which maintains the borrower's semantic information and obtains a good representation by automatically extracting important low‐ and high‐level features simultaneously. We predict the repayment of the borrower by learning discriminative features depending on the loan status. Experimental results on the Lending Club dataset show that our model is more effective than other methods. A fivefold cross‐validation is performed to run the experiments.

MicroRNAs (miRNAs) are known as an important indicator of cancers. The presence of cancer can be detected by identifying the responsible miRNAs. A fuzzy-rough entropy measure (FREM) is developed which can rank the miRNAs and thereby identify the relevant ones. FREM is used to determine the relevance of a miRNA in terms of separability between normal and cancer classes. While computing the FREM for a miRNA, fuzziness takes care of the overlapping between normal and cancer expressions, whereas rough lower approximation determines their class sizes. MiRNAs are sorted according to the highest relevance (i.e., the capability of class separation) and a percentage among them is selected from the top ranked ones. FREM is also used to determine the redundancy between two miRNAs and the redundant ones are removed from the selected set, as per the necessity. A histogram based patient selection method is also developed which can help to reduce the number of patients to be dealt during the computation of FREM, while compromising very little with the performance of the selected miRNAs for most of the data sets. The superiority of the FREM as compared to some existing methods is demonstrated extensively on six data sets in terms of sensitivity, specificity, and score. While for these data sets the score of the miRNAs selected by our method varies from 0.70 to 0.91 using SVM, those results vary from 0.37 to 0.90 for some other methods. Moreover, all the selected miRNAs corroborate with the findings of biological investigations or pathway analysis tools. The source code of FREM is available at http://www.jayanta.droppages.com/FREM.html.

Dealing with distributed data is one of the challenges for clustering, as most clustering techniques require the data to be centralized. One of them, k-means, has been elected as one of the most influential data mining algorithms for being simple, scalable, and easily modifiable to a variety of contexts and application domains. However, exact distributed versions of k-means are still sensitive to the selection of the initial cluster prototypes and require the number of clusters to be specified in advance. Additionally, preserving data privacy among repositories may be a complicating factor. In order to overcome k-means limitations, two different approaches were adopted in this paper: the first obtains a final model identical to the centralized version of the clustering algorithm and the second generates and selects clusters for each distributed data subset and combines them afterwards. It is also described how to apply the algorithms compared while preserving data privacy. The algorithms are compared experimentally from two perspectives: the theoretical one, through asymptotic complexity analyses, and the experimental one, through a comparative evaluation of results obtained from a collection of experiments and statistical tests. The results obtained indicate which algorithm is more suitable for each application scenario.

A humanoid robot has uncertain sensor data, and produces larger errors arising from the control process than other types of robot having wheels. The humanoid robot for providing service should not only generate proper behaviors to accomplish goals in unstable environments, but also consider flexible scalability to reflect demanding user's requests. We propose a control system based on planning-driven behavior selection network so as to generate autonomous behaviors of the robot rapidly and suitably. In this paper, the behavior selection network, one of behavior-based methods, is modularized considering sub-goals. The STRIPS planning makes a sequence of robot behaviors automatically by controlling the modules. The proposed system can control the robot to cope with the various environments, as well as to achieve goals according to the user's demand. Moreover, since the BSN and STRIPS planning structures are internally independent, the proposed system is scalable flexibly to increasing user requests. We confirm the usability of the proposed system by performing several test scenarios with NAO robot. Experiments show that the proposed system is able to make a behavioral sequence to fulfill goals appropriately, and can create behaviors of the robot in the various situations. We can also confirm an accuracy of 85.7% through applying the proposed system in the real world.

In this paper, we propose an automatic image enhancement tool for smart phone by using interactive differential evolution (IDE). From a remarkable progress of the camera sensor in mobile devices, people take pictures with their mobile phone instead of a digital camera. However, as they are not satisfied with their images in spite of the progress, they still want to edit their images by using mobile applications, which are usually complex and cause user fatigue, especially for beginners. To reduce it and make a simple interface, we exploit IDE, which is a kind of interactive evolutionary computation. Let the user IDE to evaluate the individuals. Because of the small parameters of the differential evolution (DE), we could make the tool simply and overcome the user fatigue. DE is also an efficient and fast evolutionary algorithm which uses the difference of the vectors. Subjective test shows the usefulness of the tool.

As digital media technologies have improved, a large amount of media content has been produced. Tagging is an effective way to manage a great volume of multimedia content. However, manual tagging has limitations such as human fatigue and subjective and ambiguous keywords. In this paper, we present an automatic tagging method to generate semantic annotation on a mobile phone. In order to overcome the constraints of the mobile environment, the method uses two layered Bayesian networks. In contrast to existing techniques, this approach attempts to design probabilistic models with fixed tree structures and intermediate nodes. To evaluate the performance of this method, an experiment is conducted with data collected over a month. The result shows the effectiveness of our proposed method. Furthermore, a simple graphic user interface is developed to visualize and evaluate recognized activities and probabilities.

Recently, Bayesian network has been widely used to cope with the uncertainty of real world in the field of artificial intelligence. Dynamic Bayesian network, a kind of Bayesian network, can solve problems in dynamic environments. However, as node and state values of node in Bayesian network grow, it is very difficult to define structure and parameter of Bayesian network. This paper proposes a method which generates and evolves structure of dynamic Bayesian network to deal with uncertainty and dynamic properties in real world using genetic algorithm. Effectiveness of the generated structure of dynamic Bayesian network is evaluated in terms of evolution process and the accuracy in a domain of the traffic accident detection.

Genetic algorithms (GA) are one of effective approaches to solve the traveling salesman problem (TSP). When applying GA to the TSP, it is necessary to use a large number of individuals in order to increase the chance of finding optimal solutions. However, this incurs high evaluation costs which make it difficult to obtain fitness values of all the individuals. To overcome this limitation we propose an efficient genetic algorithm based on fuzzy clustering which reduces evaluation costs with minimizing loss of performance. It works by evaluating only one representative individual for each cluster of a given population, and estimating the fitness values of the others from the representatives indirectly. A fuzzy c-means algorithm is used for grouping the individuals and the fitness of each individual is estimated according to membership values. The experiments were conducted with randomly generated cities, and the performance of the method was evaluated by comparing to other GAs. The results showed the usefulness of the proposed method on the TSP.

Activity recognition with mobile sensors is a challenging task due to the inherent noisy nature of the input data and resource limitations of the target platform. This paper presents a novel method of hybridizing classifier selection and classifier fusion in order to address these difficulties. It efficiently decreases the computational cost by activating appropriate classifiers according to the characteristics of the given input, and resolves the pattern variations by combining the chosen classifiers with localized templates. The proposed method is integrated with a wearable system that includes five motion sensors (accelerometers and gyroscopes), a set of bio-signal sensors, and data-gloves. The experiments on two different levels of activities, such as 11 primitive motions and eight composite behaviors, demonstrated that the proposed method is useful to the wearable systems.

Recently, inferring or sharing of mobile contexts has been actively investigated as cell phones have become more than a communication device. However, most of them focused on utilizing the contexts on social network services, while the means in mining or managing the human network itself were barely considered. In this paper, the SmartPhonebook, which mines users' social connections to manage their relationships by reasoning social and personal contexts, is presented. It works like an artificial assistant which recommends the candidate callees whom the users probably would like to contact in a certain situation. Moreover, it visualizes their social contexts like closeness and relationship with others in order to let the users know their social situations. The proposed method infers the social contexts based on the contact patterns, while it extracts the personal contexts such as the users' emotional states and behaviors from the mobile logs. Here, Bayesian networks are exploited to handle the uncertainties in the mobile environment. The proposed system has been implemented with the MS Windows Mobile 2003 SE Platform on Samsung SPH-M4650 smartphone and has been tested on real-world data. The experimental results showed that the system provides an efficient and informative way for mobile social networking.

Due to developing mobile devices and providing services like mobile blogs, people can easily share their thought and experience, at any place and any time. A picture is an important datum to record and share their thought and experience, while we can easily take pictures with a mobile device that has a camera in it. However, the quality is usually poor without image refinement. Many mobile devices provide a simply interface to improve the quality, but require knowledge of predefined filters or image enhancement to control the parameters. It causes the user to feel inconvenient in mobile environments for their real-time editing pictures. In this paper, we propose a novel image enhancement interface in consideration of the accessibility to the mobile environment and various constraints. A usability test with various images has been conducted to show its usefulness, and the proposed interface achieved better performance than the other through the SUS test.

A genetic algorithm is developed to find an optimal solution in a large search space using selection, crossover, and mutation. Some researchers have studied techniques for analysis of evolution process in genetic algorithm. In most cases, they were applied to only simple problem or they used schema theorem and numerical statistics to examine the process. These techniques are mostly developed because tracing schemas and interpreting semantics of the schemas require much effort and time. In this paper, we propose modular encoding of gene, which is used to facilitate the interpretation of the gene, identification of important parts of genetic code using evolutionary activity statistics, and analysis of the schemas. Also, we show the feasibility of the proposed method by tracing the evolution process of fuzzy robot controller.

Multi-agent based approach is reasonable and expandable methodology for simulating social phenomenon. In this paper, we simulate a developmental aspect of international trade market using multi-agent based modeling. Evolutionary computation is used to construct adaptive agent model which continuously evolves its trading strategy. Through the experiments, we show that our simulation model effectively reflects tendency of real-world international trade and existing economic research. Particularly, we find interesting result from the analysis of prevalent strategy of agents, and developmental aspects in international trade market.

A genetic algorithm can be applied to various search or optimization problems. However, there exists a problem that it takes too much cost to evaluate a large number of individuals. To deal with the problem, the fitness approximation method which reduces the cost of the evaluation with the similar performance to the general GA is needed. We proposed the fitness approximation using a combination of the approximation model and the fuzzy clustering technique. There exist two advantages of the proposed method. First, it reduces the cost of the fitness evaluation. Second, it shows the similar performance to the general GA. To verify the performance of the method, we designed the experiments using several benchmark functions and compared other fitness approximation methods.

Fingerprint classification reduces the searching time of an automated fingerprint identification system. Since fingerprints have properties of intra-class diversities and inter-class similarities, the ambiguous example causes a difficult problem in the fingerprint classification. In order to addres s the problem, we have analyzed fingerprints' subclasses with multiple decision templates. It clusters the soft outputs of support vector machines (SVMs) into several sub-classes using the self-organizing maps, and estimates a localized template for each sub-class. For an input fingerprint, the proposed method matches the output vector of SVMs to each template and finally categorizes the sample into the class of the most similar template. Experimental results on the FingerCode dataset demonstrate the effectiveness of the subclass-based approach compared with previous methods.

In a real-world, labor market consist of employer and employee, and these individuals form relationship through mutual interactions. This paper mainly focuses on development of multi-agent based evolutionary labor market by using co-evolutionary computation and game theory. Co-evolutionary computation is used to define strategy of each agent dynamically, and game theory is used for modeling relationship between employee and employer. Gift exchange game is selected as game model regard to feature of proposed labor market framework. Various experiments were performed, and we analyzed the variation of interactions between employee and employer. Through the experimental result, we concluded that balanced power between employee and employer is important factor in maintenance and extension of labor market.

Weblog is one of the most spread web services. The content of the weblog includes daily events and emotions. If we collect personal information using mobile devices and create a weblog, user can create their own weblog easily. Some researchers already developed systems that created weblog in mobile environment. In this paper, user's activity is inferred from personal information in mobile device. The inferred activities and story generation engine are used to generate text for creating a weblog. Finally, the text, photographs and user's movement in Google Map are integrated into a weblog.

Image enhancement is an important issue in digital image processing. Various approaches have been developed to solve image enhancement problems, but most of them require deep expert knowledge to design appropriate image filters. To automatically design a filter, we propose a novel approach based on the genetic algorithm that optimizes a set of standard filters by determining their types and order. Moreover, the proposed method is able to manage various types of noise factors. We applied the proposed method to local and global image enhancement problems such as impulsive noise reduction, interpolation, and orientation enhancement. In terms of subjective and objective evaluations, the results show the superiority of the proposed method.

Gene selection is an important issue for cancer classification based on gene expression profiles. Filter and wrapper approaches are used widely for gene selection, where the former is hard to measure the relationship between genes and the latter requires lots of computation. We present a novel method, called gene boosting, to select relevant gene subsets by integrating filter and wrapper approaches. It repeatedly selects a set of top-ranked informative genes by a filtering algorithm with respect to a temporal training dataset constructed according to the classification result for the original training dataset. Empirical results on three microarray benchmark datasets have shown that the proposed method is effective and efficient in finding a relevant and concise gene subset. Competitive performance was achieved with fewer genes in a reasonable time. This also led to the identification of some genes selected frequently as useful features.

BACKGROUND: For the past few years, scientific controversy has surrounded the large number of errors in forensic and literature mitochondrial DNA (mtDNA) data. However, recent research has shown that using mtDNA phylogeny and referring to known mtDNA haplotypes can be useful for checking the quality of sequence data. RESULTS: We developed a Web-based bioinformatics resource "mtDNAmanager" that offers a convenient interface supporting the management and quality analysis of mtDNA sequence data. The mtDNAmanager performs computations on mtDNA control-region sequences to estimate the most-probable mtDNA haplogroups and retrieves similar sequences from a selected database. By the phased designation of the most-probable haplogroups (both expected and estimated haplogroups), mtDNAmanager enables users to systematically detect errors whilst allowing for confirmation of the presence of clear key diagnostic mutations and accompanying mutations. The query tools of mtDNAmanager also facilitate database screening with two options of "match" and "include the queried nucleotide polymorphism". In addition, mtDNAmanager provides Web interfaces for users to manage and analyse their own data in batch mode. CONCLUSION: The mtDNAmanager will provide systematic routines for mtDNA sequence data management and analysis via easily accessible Web interfaces, and thus should be very useful for population, medical and forensic studies that employ mtDNA analysis. mtDNAmanager can be accessed at http://mtmanager.yonsei.ac.kr.

In general, the analysis of microarray data requires two steps: feature selection and classification. From a variety of feature selection methods and classifiers, it is difficult to find optimal ensembles composed of any feature-classifier pairs. This paper proposes a novel method based on the evolutionary algorithm (EA) to form sophisticated ensembles of features and classifiers that can be used to obtain high classification performance. In spite of the exponential number of possible ensembles of individual feature-classifier pairs, an EA can produce the best ensemble in a reasonable amount of time. The chromosome is encoded with real values to decide the weight for each feature-classifier pair in an ensemble. Experimental results with two well-known microarray datasets in terms of time and classification rate indicate that the proposed method produces ensembles that are superior to individual classifiers, as well as other ensembles optimized by random and greedy strategies.

This paper proposes a novel fingerprint classification method using multiple decision templates of Support Vector Machines (SVMs) with adaptive features. In order to overcome intra-class and inter-class ambiguities of fingerprints, the proposed method extracts a feature vector from an adaptively detected feature region and classifies the feature vector using SVMs. The outputs of the SVMs are then combined by multiple decision templates that make several per class. Experimental results on NIST4 fingerprint database revealed the effectiveness and validity of the proposed method for fingerprint classification.

In recent years, the concept of “autonomous mental development” (AMD) has been applied to the construction of artificial systems such as conversational agents, in order to resolve some of the difficulties involved in the manual definition of their knowledge bases and behavioral patterns. AMD is a new paradigm for developing autonomous machines, which are adaptive and flexible to the environment. Language development, a kind of mental development, is an important aspect of intelligent conversational agents. In this paper, we propose an intelligent conversational agent and its language development mechanism by putting together five promising techniques: Bayesian networks, pattern matching, finite-state machines, templates, and genetic programming (GP). Knowledge acquisition implemented by finite-state machines and templates, and language learning by GP are used for language development. Several illustrations and usability tests show the usefulness of the proposed developmental conversational agent.

As the usage of service robots becomes more sophisticated, direct communication by means of human language is required to increase the efficiency of their performance. In natural speech interaction, however, people often omit some words and rely on background knowledge or the context, resulting in ambiguity. In order to develop smarter service robots, therefore, managing the context of interaction is essential. In this correspondence, we have investigated the mixed-initiative interaction that prompts for missing information and clarifies ambiguous statements based on hierarchically designed Bayesian networks. Simulation with the Kephera II robot and a usability test have demonstrated the usefulness of the proposed method.

The evolutionary approach for gaming is different from the traditional one that exploits knowledge of the opening, middle, and endgame stages. It is therefore sometimes inefficient to evolve simple heuristics that may be created easily by humans because it is based purely on a bottom-up style of construction. Incorporating domain knowledge into evolutionary computation can improve the performance of evolved strategies and accelerate the speed of evolution by reducing the search space. In this paper, we develop an evolutionary Othello player with the systematic insertion of opening knowledge into the framework of evolution. The probability of opening selection is coming from the expert’s opening list. Preliminary experimental results show that the proposed method is promising for generating better strategies for Othello players.

Bayesian network (BN) is a useful tool to represent joint probability distribution in the form of graphical model providing flexible inference and uncertainty handling. If there is enough knowledge about domain, it is possible to design the structure and parameters of BN by expert. Also, it can be learned from massive dataset with statistical learning algorithm. Usually, because the search space of Bayesian networks is relatively huge compared to the other models, evolutionary algorithms have been used to find optimal structure and parameters by many researchers. In this paper, we have focused on the topic of adaptation of constructed models for better performance. If there are a number of models constructed or learned by different experts or sources, it is better to fuse them into one model by considering all the information of each model. However, the complexity of the integrated model is relatively higher than previous isolated models. Minimizing the complexity of the integrated model using evolutionary algorithm is proposed. After integrating models into single one, it needs to adapt to the new data from the environment. It is likely to provide wrong results to the newly generated data from the environment and slightly modifying the joint probability distribution is necessary. The refinement process is also guided by the evolutionary algorithm because the space of search is large. Experimental results on a benchmark network show that the proposed adaptation methods with evolutionary algorithm can perform better than heuristics or greedy approaches.

We review the applications of artificial life (ALife), the creation of synthetic life on computers to study, simulate, and understand living systems. The definition and features of ALife are shown by application studies. ALife application fields treated include robot control, robot manufacturing, practical robots, computer graphics, natural phenomenon modeling, entertainment, games, music, economics, Internet, information processing, industrial design, simulation software, electronics, security, data mining, and telecommunications. In order to show the status of ALife application research, this review primarily features a survey of about 180 ALife application articles rather than a selected representation of a few articles. Evolutionary computation is the most popular method for designing such applications, but recently swarm intelligence, artificial immune network, and agent-based modeling have also produced results. Applications were initially restricted to the robotics and computer graphics, but presently, many different applications in engineering areas are of interest.

The process of learning the behavior of a given program by using machine-learning techniques (based on system-call audit data) is effective to detect intrusions. Rule learning, neural networks, statistics, and hidden Markov models (HMMs) are some of the kinds of representative methods for intrusion detection. Among them, neural networks are known for good performance in learning system-call sequences. In order to apply this knowledge to real-world problems successfully, it is important to determine the structures and weights of these call sequences. However, finding the appropriate structures requires very long time periods because there are no suitable analytical solutions. In this paper, a novel intrusion-detection technique based on evolutionary neural networks (ENNs) is proposed. One advantage of using ENNs is that it takes less time to obtain superior neural networks than when using conventional approaches. This is because they discover the structures and weights of the neural networks simultaneously. Experimental results with the 1999 Defense Advanced Research Projects Agency (DARPA) Intrusion Detection Evaluation (IDEVAL) data confirm that ENNs are promising tools for intrusion detection.

The service robot supports people in their daily activities, while the interaction between humans and robots seems to be an important part of its performance. Dialogue may be beneficial to the robot to increase the flexibility and facility of the interaction. Traditional robots have merely dealt with simple queries like commands, but in conversation people often omit some words because of the background knowledge or the context of the conversation. Since environments contain various uncertainties, managing the context of a dialogue or the uncertainties should be necessary to support smarter service robots. In order to establish a natural communication between people and robots, we have been investigating the use of mixed-initiative interaction that prompts for missing concepts and clarifies for spurious concepts. Hierarchically designed Bayesian networks are presented for the mixed-initiative interaction. A simulation and a real robot are constructed for the demonstration of the proposed method, and experiments also show the usefulness.

The evolutionary approach for gaming is different from the traditional one that exploits knowledge of the opening, middle, and endgame stages. It is, therefore, sometimes inefficient to evolve simple heuristics that may be created easily by humans because it is based purely on a bottom-up style of construction. Incorporating domain knowledge into evolutionary computation can improve the performance of evolved strategies and accelerate the speed of evolution by reducing the search space. In this paper, we propose the systematic insertion of opening knowledge and an endgame database into the framework of evolutionary checkers. Also, the common knowledge that the combination of diverse strategies is better than a single best one is included in the middle stage and is implemented using crowding algorithm and a strategy combination scheme. Experimental results show that the proposed method is promising for generating better strategies.

With the development of microarray technology, the classification of microarray data has risen as an important topic over the past decade. From various feature selection methods and classifiers, it is very hard to find a perfect method to classify microarray data due to the incompleteness of algorithms, the defects of data, etc. This paper proposes a sophisticated ensemble of such features and classifiers to obtain high classification performance. Speciated genetic algorithm has been exploited to get the diverse ensembles of features and classifiers in a reasonable time. Experimental results with two well-known datasets indicate that the proposed method finds many good ensembles that are superior to other individual classifiers.

Various digital characters, which are automatic and intelligent, are attempted with the introduction of artificial intelligence or artificial life. Since a character's behavior is designed by a developer, the style can be static and simple. Even complex patterns designed by a developer cannot satisfy various users and easily make them feel tedious. A game should maintain various and complex character's behaviors, but it is not easy for the developer to design them. In this paper, we adopt genetic algorithm to produce various and excellent behavior-styles for characters especially focusing on Robocode which is one of the promising simulators for artificial intelligence.

Clustering of gene expression profiles has been used for gene function identification. Since the genes usually belong to multiple functional families, fuzzy clustering methods are appropriate. However, a natural way to measure the quality of the fuzzy cluster partitions is still required. A Bayesian validation method for fuzzy partition selection with the largest posterior probability given the dataset is proposed. This method is compared to four representative fuzzy cluster validity measures using fuzzy c-means algorithm on four well-known datasets in terms of the number of clusters predicted in the data. An analysis of Saccharomyces cerevisiae cell cycle gene expression data follows to show the usefulness of the proposed method.

Several techniques in artificial intelligence have shown a great potential to develop useful human-computer interfaces, but it is still quite far from realizing a system of matching the human performance, especially in terms of emotion, intuition and inspiration. To overcome this shortcoming, we present a promising technique called interactive genetic algorithm (IGA), which performs optimization with human evaluation, and with which the user can obtain what he has in mind through repeated interaction. To project the usefulness of the IGA to develop emotional human-computer interfaces, we have applied it to the problems of image and music information retrieval. Several experiments show that our approach allows us to design and search digital media not only explicitly expressed, but also abstract images such as "cheerful impression," and "gloomy impression." It is expected that the same approach can be applied to many other problems in musical information retrieval and manipulation based on intuition and inspiration.

SGML/XML are having a profound impact on data modeling and processing. We present an efficient algorithm to compute differences between old and new versions of an SGML/XML document. The difference between the two versions can be considered to be an edit script that transforms one document tree into another. The proposed algorithm is based on a hybridization of bottom-up and top-down methods: The matching relationships between nodes in the two versions are produced in a bottom-up manner and then the top-down breadth-first search computes an edit script. Faster matching is achieved because the algorithm does not need to investigate the possible existence of matchings for all nodes. Furthermore, it can detect structurally meaningful changes such as the movement and copy of a subtree as well as simple changes to the node itself like insertion, deletion, and update.

Evolvable hardware adjusts oneself to changeable environments by self-organizing the circuit. Due to its high productivity and creativity for designing circuit, it is widely investigated. However, it is very difficult to apply it to a complicated circuit, because the search space increases exponentially as the complexity of hardware. In this paper, we propose a modular approach to evolving complex hardware circuits effectively. A comparative experiment with the conventional evolutionary approach indicates that the proposed method works 50/spl sim/1000 times faster and yields a more optimized hardware.

DNA microarray technology provides a format for the simultaneous measurement of the expression level of thousands of genes in a single hybridization assay. One exciting result of microarray technology has been the demonstration that patterns of gene expression can distinguish between tumors of different anatomical origins. Standard statistical methodologies in classification and prediction do not work well or even at all when N (the number of samples) < p (genes). Modification of existing statistical methodologies or development of new methodologies are needed for the analysis of cancer. Recently, designing artificial neural networks (ANNs) by evolutionary algorithms has emerged as a preferred alternative to the common practice of selecting the apparent best network. We propose an evolutionary neural network that classifies gene expression profiles into normal or colon cancer cell. Colon cancer is the second only to lung cancer as a cause of cancer-related mortality in Western countries. Colon cancer is a genetic disease, propagated by the acquisition of somatic alterations that influence gene expression. Experimental results on colon microarray data with evolutionary neural network show that the proposed method can perform better than other classifiers. Contribution of this article is applying evolutionary neural network to gene expression classification problem.

Cancer is one of the most dangerous diseases for people. Recently, development of array technologies makes it possible to measure thousands of genes at once, and it can be used to treat cancer. Various methods using array data to classify cancer are proposed, but there are no perfect and general methods. Ensemble method can demonstrate its ability if there are complementary set of individuals. However, it is needed methods to search the optimal ensembles, because there are so many available ensembles. We propose a GA-based method to search the optimal ensemble. In two benchmark datasets, our proposed method has shown significant result in the aspects of performance and time efficiency under the several situations.

Recently, many researchers attempt to develop an effective SOM-based pattern recognizer for high performance classification. Structure adaptive self-organizing map (SASOM) is a variant of SOM that is useful to pattern recognition and visualization. Fusion of classifiers can overcome the limitation of a single classifier by complementing each other. Fuzzy integral is a combination scheme that uses subjectively defined relevance of classifiers. In this paper, fusion of SASOM's using fuzzy integral is proposed for Web mining problem. User profile represents different aspects of user's characteristics and needs an ensemble of classifier that estimate user's preference using Web content labeled by user as "like" or "dislike." The proposed method estimates the user profile using subsets of important features extracted from user-rated Web documents. Using UCI Syskill & Webert data, the method is tested and compared with other classifier including ID3, BP and naive Bayes classifier. Experimental results show that the fusion of SASOM's using fuzzy integral can perform better than not only previous studies but also majority voting of SASOM's.

Gene expression profile is numerical data of gene expression levels from organism, measured on the microarray. In general, each specific tissue indicates different expression level in related genes, so that it is possible to classify disease by gene expression profile. For classification, it is needed to select related genes called feature selection, because all the genes are not useful for classification. We propose GA-based method for searching optimal ensemble of feature-classifier pairs of gene expression profile in seven feature selection methods based on correlation, distance, and information theory, and representative six classifiers. Experimental results on two gene expression profiles related to cancers show that GA finds good solution quickly. Especially, in Lymphoma dataset, GA finds the ensemble of 100% accuracy.

While several conventional techniques for diagnosis of cancer in clinical practice can be often incomplete or misleading, molecular level diagnostics with gene expression profiles can offer the methodology of precise, objective, and systematic cancer classification. Moreover, since accurate classification of cancer is very important issue for treatment of cancer, it is desirable to make a decision by combining the results of various basis classifiers rather than by deciding the result with only one classifier. Generally combining classifiers gives high performance and high confidence. In spite of many advantages of ensemble classifiers, ensemble with mutually error-correlated classifiers has a limit in the performance. In this paper, we propose the ensemble of neural network classifiers learned from negatively correlated features to precisely classify cancer, and systematically evaluate the performances of the proposed method using three benchmark datasets. Experimental results show that the ensemble classifier with negatively correlated features produces the best recognition rate on the three benchmark datasets.

As the reliance on computers increases, security of critical computers becomes more important. An IDS detects unauthorized usage and misuse by a local user as well as modification of important data by analyzing system calls, system logs, activation time, and network packets Conventional IDSs based on anomaly detection employ several artificial intelligence techniques to model normal behavior. However, they have the shortcoming that there are undetectable intrusions according to types for each measure and modeling method because each intrusion type results in anomalies. We propose a multiple-measure intrusion detection method to remedy this drawback of conventional anomaly detectors. We measure normal behavior by system calls, resource usage and file access events and build up profiles for normal behavior with a hidden Markov model, statistical method and rule-base method, which are integrated with a rule-based approach. Experimental results with real data clearly demonstrate the effectiveness of the proposed method that has a significantly low false-positive error rate against various types of intrusion.

Microarray technology has supplied a large volume of data, which changes many problems in biology into the problems of computing. As a result techniques for extracting useful information from the data are developed. In particular, microarray technology has been applied to prediction and diagnosis of cancer, so that it expectedly helps us to exactly predict and diagnose cancer. To precisely classify cancer we have to select genes related to cancer because the genes extracted from microarray have many noises. In this paper, we attempt to explore seven feature selection methods and four classifiers and propose ensemble classifiers in three benchmark datasets to systematically evaluate the performances of the feature selection methods and machine learning classifiers. Three benchmark datasets are leukemia cancer dataset, colon cancer dataset and lymphoma cancer data set. The methods to combine the classifiers are majority voting, weighted voting, and Bayesian approach to improve the performance of classification. Experimental results show that the ensemble with several basis classifiers produces the best recognition rate on the benchmark datasets.

This paper presents a syntactic method for sophisticated logical structure analysis that transforms document images with multiple pages and hierarchical structure into an electronic document based on SGML/XML. To produce a logical structure more accurately and quickly than previous works of which the basic units are text lines, the proposed parsing method takes text regions with hierarchical structure as input. Furthermore, we define a document model that is able to describe geometric characteristics and logical structure information of documents efficiently and present its automated creation method. Experimental results with 372 images scanned from the IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) show that the method has performed logical structure analysis successfully and generated a document model automatically. Particularly, the method generates SGML/XML documents as the result of structural analysis, so that it enhances the reusability of documents and independence of platform.

Evolvable hardware (EHW) has become an attractive topic recently because such hardware can reconfigure itself to adapt to the environment it is embedded in. EHW uses a genetic algorithm (GA), which is one of the evolutionary algorithms, to search for the goal hardware. In this paper, we propose EHW using a speciated GA that can evolve diverse circuits with single-step evolution. The speciation algorithm helps to find diverse solutions as the result of the evolution, and maintains the diversity during the evolution. We have applied a fitness-sharing method for speciation to the EHW of a 6-multiplexer, and have obtained diverse hardware structures. Also, we have found a circuit in 35% less generations than we did with a conventional genetic algorithm.

Conventional evolutionary algorithms have a property that only one solution often dominates and it is sometimes useful to find diverse solutions and combine them because there might be many different solutions to one problem in real-world problems. Recently, developing checkers players using evolutionary algorithms has been widely exploited to show the power of evolution for machine learning. In this paper, we propose an evolutionary checkers player that is developed by a speciation technique called the "crowding algorithm". In many experiments, our checkers player with an ensemble structure showed better performance than non-speciated checkers players. A neural network is used to validate the game board, and a min-max search finds the optimal board. The neural network evaluator is evolved using the evolutionary algorithm.

This paper illustrates the evolutionary adaptive process of the rules of a fuzzy controller evolved by a genetic algorithm. Evolutionary activity and schema analysis are used to evaluate and analyze the evolution. The analysis shows that the evolution has been adaptive and final fuzzy rules have evolved from adaptive ones of earlier generations.

Interactive evolutionary computation (IEC), which takes a user's evaluation as a fitness function, performs poorly for local search due to the limitation of population size and generation length. To solve this, the direct manipulation (DM) method, well known in HCI, of evolution for IEC has been proposed. It allows the user to manipulate individuals directly, instead of using evolutionary operators as an interface to each individual. In this paper, we analyze the usefulness of DM with a fitness landscape and N-K model. We have applied the DM concept to a fashion design system based on IEC, and analyzed the results with the concept of fitness landscape and Boolean hypercube.

Bioinformatics has recently drawn a lot of attention to efficiently analyze biological genomic information with information technology, especially pattern recognition. In this paper, we attempt to explore extensive features and classifiers through a comparative study of the most promising feature selection methods and machine learning classifiers. The gene information from a patient's marrow expressed by DNA microarray, which is either the acute myeloid leukemia or acute lymphoblastic leukemia, is used to predict the cancer class. Pearson's and Spearman's correlation coefficients, Euclidean distance, cosine coefficient, information gain, mutual information and signal to noise ratio have been used for feature selection. Backpropagation neural network, self-organizing map, structure adaptive self-organizing map, support vector machine, inductive decision tree and k-nearest neighbor have been used for classification. Experimental results indicate that backpropagation neural network with Pearson's correlation coefficients produces the best result, 97.1% of recognition rate on the test data.

The explosion of DNA and protein sequence data in public and private databases has been encouraging interdisciplinary research on biology and information technology. Gene expression profiles are just sequences of numbers, and the necessity of tools analyzing them to get useful information has risen significantly. In order to predict the cancer class of patients from the gene expression profile, this paper presents a classification framework that combines a pair of classifiers trained with mutually exclusive features. The idea behind feature selection with nonoverlapping correlation is to encourage classifier ensemble, which consists of multiple classifiers, to learn different aspects of training data, so that classifiers can search in a wide solution space. Experimental results show that the classifier ensemble produces higher recognition accuracy than conventional classifiers.

There are a lot of industrial applications that can be solved competitively by hard computing, while still requiring the tolerance for imprecision and uncertainty that can be exploited by soft computing. This paper presents a novel intrusion detection system (IDS) that models normal behaviors with hidden Markov models (HMM) and attempts to detect intrusions by noting significant deviations from the models. Among several soft computing techniques neural network and fuzzy logic are incorporated into the system to achieve robustness and flexibility. The self-organizing map (SOM) determines the optimal measures of audit data and reduces them into appropriate size for efficient modeling by HMM. Based on several models with different measures, fuzzy logic makes the final decision of whether current behavior is abnormal or not. Experimental results with some real audit data show that the proposed fusion produces a viable intrusion detection system. Fuzzy rules that utilize the models based on the measures of system call, file access, and the combination of them produce more reliable performance.

Content-based image retrieval has been actively studied in several fields. This provides more effective management and retrieval of images than the keyword-based approach. However, most of the conventional methods lack the capability to effectively incorporate human intuition and emotion into retrieving images. It is difficult to obtain satisfactory results when the user wants the image that cannot be explicitly described or can be requested only based on impression. In order to solve this problem and supplement the lack of the user's expression capability, we have developed an image retrieval system based on human preference and emotion by using an interactive genetic algorithm (IGA). This system extracts the feature from images by wavelet transform, and provides a user-friendly means to retrieve an image from a large database when the user cannot clearly define what the image must be. Therefore, this facilitates the search for the image not only with explicit queries, but also with implicit queries such as "cheerful impression," "gloomy impression," and so on. A thorough experiment with a 2000 image database shows the usefulness of the proposed system.

In order to develop effective evolutionary artificial neural networks (EANNs) we have to address the questions on how to evolve EANNs more efficiently and how to achieve the best performance from the ANNs evolved. Most of the previous works, however, do not utilize all the information obtained with several ANNs but choose the one best network in the last generation. Some recent works indicate that making use of population information by combining ANNs in the last generation can improve the performance, because they can complement each other to construct effective multiple neural networks. We propose a new method of evolving multiple speciated neural networks by fitness sharing which helps to optimize multi-objective functions with genetic algorithms. Experiments with the breast cancer data from UCI benchmark datasets show that the proposed method can produce more speciated ANNs and improve the performance by combining the only representative individuals.

To solve a general problem with genetic algorithms, it is desirable to maintain the population size as large as possible. In some cases, however, the cost to evaluate each individual is relatively high, and it is difficult to maintain a large population. To solve this problem, we propose a hybrid GA based on clustering, which considerably reduces the evaluation number without any loss of performance. The algorithm divides the whole population into several clusters, and evaluates only one representative for each cluster. The fitness values of other individuals are estimated from the representative fitness values indirectly, which can maintain a large population with less number of evaluations. Several benchmark tests have been conducted and the results show that the proposed GA is very efficient.

In behavior-based robotics the control of a robot is shared between a set of purposive perception-action units, called behaviors. A major issue in the design of behavior-based control systems is the formulation of effective mechanisms for coordination of the behaviors' activities into strategies for rational and coherent behavior. There has been extensive work to construct an optimal controller for a mobile robot by evolutionary approaches such as genetic algorithm, genetic programming, and so on. In this line of research, we have also presented a method of applying CAM-Brain, evolved neural networks based on cellular automata (CA), to control a mobile robot. However, this approach has limitations to make the robot to perform appropriate behavior in complex environments. The multi module coordination method can make complex and general behaviors by combining several modules evolved or programmed, to do a simple behavior. In this paper, we coordinate several modules evolved to do a simple behavior by Maes's action selection mechanism. Maes (1989) has proposed a mechanism for action selection, which is reviewed here and is evaluated using a simulation environment. Experimental results show that this approach has potential to develop a sophisticated evolutionary neural controller for complex environments.

Explaining emergence is a difficult work, such that there are many arguments on what it is or how it can be explained. Nonetheless, it is frequently referred to in many fields, such as behavior-based robotics, artificial life and complex systems, without any formal definition. In this paper, we develop a fuzzy logic controller for a simulated mobile robot with a genetic algorithm and analyze the behavior of the controller from the perspective of observational emergence. The analysis shows that the fuzzy logic controller has acquired emergent behavior through the interactions of the underlying fuzzy rules.

Recently, there has been extensive work on the construction of fuzzy controllers for mobile robots by a genetic algorithm (GA); therefore, we can realize evolutionary optimization as a promising method for developing fuzzy controllers. However, much investigation on the evolutionary fuzzy controller remains because most of the previous works have not seriously attempted to analyze the fuzzy controller obtained by evolution. This paper develops a fuzzy logic controller for a mobile robot with a GA in simulation environments and analyzes the behaviors of the controller with a state transition diagram of the internal model. Experimental results show that appropriate control mechanisms of the fuzzy controller are obtained by evolution. The controller has evolved wen enough to smoothly drive the robot in different environments. The robot produces emergent behaviors by the interaction of several fuzzy rules obtained.

Adaptive behaviors often emerge through interactions between adjacent neighbors in dynamic systems, such as social and economic systems. In many cases, an individual's behavior can be modeled by a stimulus-response system in a dynamic environment. In this paper, we use the iterated prisoner's dilemma (IPD) game, which is simple yet capable of dealing with complex problems, to model a dynamic system such as social or economic systems. We investigate coalitions consisting of many players and their emergence in a co-evolutionary learning environment. We introduce the concept of confidence for players in a coalition and show how such confidences help to improve the generalization ability of the whole coalition. Experimental results are presented to demonstrate that co-evolutionary learning with coalitions and player confidences can produce IPD game-playing strategies that generalize well.

This paper presents a knowledge-based method for sophisticated geometric structure analysis of technical journal pages. The proposed knowledge base encodes geometric characteristics that are not only common in technical journals but also publication-specific in the form of rules. The method takes the hybrid of top-down and bottom-up techniques and consists of two phases: region segmentation and identification. Generally, the result of the segmentation process does not have a one-to-one matching with composite layout components. Therefore, the proposed method identifies non-text objects, such as images, drawings, and tables, as well as text objects, by splitting or grouping segmented regions into composite layout components. Experimental results with 372 images scanned from the IEEE Transactions on Pattern Analysis and Machine Intelligence show that the proposed method has performed geometric structure analysis successfully on more than 99 percent of the test images.

The N-player iterated prisoner's dilemma (NIPD) game has been used widely to study the evolution of cooperation in social, economic and biological systems. Previous work on the NIPD game studied the impact of the number of players and the payoff function on the evolution of cooperation. This paper studies the localization issue in the NIPD game and investigates the impact of local interaction on genetically evolved strategies for the NIPD game. Our experimental results show that localization of interaction has a major impact on the evolution of cooperative coalitions, while localized learning makes the population oscillate. This paper also investigates the effect of the history length in the NIPD game. It is found experimentally that a longer history makes a population more stable, but it takes longer time to reach this stable state.

Recently, there is a growing interest from various fields of studies in GIS which promotes efficient storage and retrieval of geographic information. The development of automated vectorizing system, an input method for GIS, is of extreme importance due to the fact that this part of constructing a GIS has been time consuming. Most vectorizing systems require users to set parameters as appropriately as possible to the map image at hand, but it is quite difficult for a novice to adjust the parameters accurately. In this paper, we propose an automated vectorizing system based on a knowledge base rendering an appropriate choice of the parameters. Each rule in the knowledge base is characterized by the type of maps, resolution, line width, slope and protrusion. Experimental results with various map images evince the proposed system to be superior in terms of performance and convenience.

There is a growing interest from various fields of studies in geographic information systems (GIS) which promotes efficient storage and retrieval of geographic information. The development of automated vectorizing system, an input method for GIS, is of extreme importance due to the fact that this part of constructing a GIS has been time consuming. Most vectorizing systems require users to set parameters as appropriately as possible to the map image at hand, but it is quite difficult for a novice to adjust the parameters accurately. We propose an automated vectorizing system based on a knowledge base rendering an appropriate choice of the parameters. Each rule in the knowledge base is characterized by the type of maps, resolution, line width, slope and protrusion. Experimental results with various map evince the proposed system to be superior in terms of performance and convenience.

The evolutionary approach to artificial neural networks has been rapidly developing in recent years and shows great potential as a powerful tool. However, most evolutionary neural networks have paid little attention to the fact that they can evolve from modules. This paper presents a hybrid method of modular neural networks and evolutionary algorithm as a promising model for intelligent systems. To build a neural network system that is rich in autonomy and creativity, some ideas of artificial life have been adopted. This paper describes the concepts and methodologies for the evolvable model of modular neural networks, which might not only develop spontaneously new functionality, but also grow and evolve its own structure autonomously. We show the potential of the method by applying it to a visual categorization task with handwritten digits. The evolutionary mechanism has shown a strong potential to generate useful network architectures from an initial set of randomly connected networks. © 1998 John Wiley & Sons, Inc.

This article presents a simple yet elegant pattern recognizer based on a dynamic node-splitting scheme for the self-organizing map that can adapt its structure as well as its weights. The scheme makes use of a structure adaptation capability to place the nodes of prototype vectors into the pattern space accurately so as to make the decision boundaries as close to the class boundaries as possible. In order to show the performance of the proposed scheme, experiments with the unconstrained handwritten digit database of Concordia University in Canada were conducted. The proposed method for an incremental formation of feature maps is 96.05 percent of the recognition rate. In view of the elegant simplicity of the approach, the reported performance is remarkable and can stand up to one of the best results reported in the literature with the same database.

Artificial neural networks have been recognized as a powerful tool for pattern classification problems, but a number of researchers have also suggested that straightforward neural-network approaches to pattern recognition are largely inadequate for difficult problems such as handwritten numeral recognition. In this paper, we present three sophisticated neural-network classifiers to solve complex pattern recognition problems: multiple multilayer perceptron (MLP) classifier, hidden Markov model (HMM)/MLP hybrid classifier, and structure-adaptive self-organizing map (SOM) classifier. In order to verify the superiority of the proposed classifiers, experiments were performed with the unconstrained handwritten numeral database of Concordia University, Montreal, Canada. The three methods have produced 97.35%, 96.55%, and 96.05% of the recognition rates, respectively, which are better than those of several previous methods reported in the literature on the same database.

In this paper we present an efficient pattern recognizer based on a self-organizing neural network which can adapt its structure as well as its weights. The network, called doubly self-organizing neural network (DSNN), makes use of the structure-adaptation capability to place the nodes of prototype vectors into the pattern space accurately so as to make the decision boundaries as close to the class boundaries as possible. In order to verify the superiority of the DSNN, experiments with the unconstrained handwritten numeral database of Concordia University in Canada were conducted. The proposed method has produced 96.05% of the recognition rate, which we show better than those of several previous methods reported in the literature on the same database.

Document recognition usually requires several preprocessing steps in which skew estimation and correction are critical to get a useful system. This paper proposes an efficient data reduction method to enhance the performance of document skew estimation by using a Hough transformation. The time complexity of the Hough transformation is O(/spl Theta/N), where N is the number of black pixels in a document and /spl Theta/ is the skew estimation range divided by /spl Delta//spl theta/. We might enhance the performance by reducing N or /spl Theta/. The proposed method uses an efficient data reduction method called the modified version of divided horizontal histograms, which reduces the number of black pixels N, while retaining the skewness of document. In order to show the superiority of the proposed method, we have also performed experiments with scanned documents, comparing the result with those of the usual data reduction methods: vertical run-length and connected component methods.

Most of the multiple recognizers system use a single combination method, therefore recognition performance depends on the characteristics of selected combination method. In order to solve this dependency problem and to increase recognition performance, we propose a new combination architecture of multiple recognizer that has two combination stages. The proposed system consists of three stages: 1) the recognition stage including 5 recognizers, 2) the first combination stage including 3 combinators which belong to different level, and 3) the combination stage including a simple combinator. We verify the performance of the proposed system using two standard handwritten digit database, CEDAR and CENPARMI, and recognition performance is better than other single combinator systems.

Efficiency is a problem in automatic programming, both in the programs produced and in the synthesis process itself. This paper presents a framework for using evolutionary mechanisms to guide program synthesis. A particular implementation of the framework, called Tierra, is described. Given a naive program and some limits on system resources, Tierra generates mutated programs and evolution proceeds by natural selection as the programs compete for Control Processing Unit (CPU) time and memory space. By applying the evolutionary mechanisms, Tierra has guided the automatic implementation of an efficient self-replicating program. The system is under continuing development; it is viewed more as a research tool than a prototype programming assistant; however, the performance of the system at present gives some hope for the ultimate feasibility of such systems.

This paper presents a hybrid architecture of hidden Markov models (HMMs) and a multilayer perceptron (MLP). This exploits the discriminative capability of a neural network classifier while using HMM formalism to capture the dynamics of input patterns. The main purpose is to improve the discriminative power of the HMM-based recognizer by additionally classifying the likelihood values inside them with an MLP classifier. To appreciate the performance of the presented method, we apply it to the recognition problem of on-line handwritten characters. Simulations show that the proposed architecture leads to a significant improvement in generalization performance over conventional approaches to sequential pattern recognition.

Multiplayer feedforward networks trained by minimizing the mean squared error and by using a one of c teaching function yield network outputs that estimate posterior class probabilities. This provides a sound basis for combining the results from multiple networks to get more accurate classification. This paper presents a method for combining multiple networks based on fuzzy logic, especially the fuzzy integral. This method non-linearly combines objective evidence, in the form of a network output, with subjective evaluation of the importance of the individual neural networks. The experimental results with the recognition problem of on-line handwriting characters show that the performance of individual networks could be improved significantly.

In the area of artificial neural networks, the concept of combining multiple networks has been proposed as a new direction for the development of highly reliable neural network systems. The authors propose a method for multinetwork combination based on the fuzzy integral. This technique nonlinearly combines objective evidence, in the form of a fuzzy membership function, with subjective evaluation of the worth of the individual neural networks with respect to the decision. The experimental results with the recognition problem of on-line handwriting characters confirm the superiority of the presented method to the other voting techniques.>

A hierarchical neural network which recognizes printed Hangul (Korean) characters is proposed. This system is composed of a type-classification network and six recognition networks. The former classifies input character images into one of the six types by their overall structure, and the latter further classify them into character code. A training scheme including systematic noises is introduced for improving the generalization capabilities of the networks. With the noise-included training, the recognition rate is up to 98.28%, which is superior to the conventional back-propagation network. The neural network approach is very reasonable compared to statistical classifiers and an analysis of generalization capability demonstrates acceptable performance