Effective data augmentation for domain-specific image classification must balance three competing objectives: diversity, faithfulness, and label clarity. However, current methods, including state-of-the-art diffusion models, struggle to achieve this balance and are further limited by issues such as stochastic outputs under strong transformations. We propose SGD-Mix, a novel framework that systematically reconciles these objectives. Our approach employs saliency-guided mixing to preserve foreground semantics while introducing diverse backgrounds, followed by a domain-specific fine-tuned diffusion model that refines the output to ensure high fidelity and strict label consistency. Extensive experiments across fine-grained, long-tail, few-shot, and background robustness tasks demonstrate that SGD-Mix achieves state-of-the-art performance, surpassing existing diffusion-based and non-generative methods by notable margins. Our code is available at https://github.com/RadonDong/sgdmix

Cone-Beam Computed Tomography (CBCT) has gar-nered significant attention due to lower radiation dosage and faster scanning time, which has been widely used in clinical applications for decades. However, its poor image quality is always challenging to clinical experts. To address this problem, we propose our work DiffuCE,$a$Diffusion model framework for CBCT Enhancement. The main contributions of our work are three-fold: (1) Increased Gen-eralizability: Our training data exclusively comprises pixel space data, eliminating the necessity for additional imaging machine settings. This emphasizes the model's ability to generalize effectively across diverse conditions. (2) Effi-cient Training: Rather than starting from scratch, our approachfine-tunesfrom a well-established foundation model. This illustrates the viability of efficient training strategies for medical image restoration tasks, optimizing resource utilization. (3) Competitive Performance: DiffuCE exhibits outstanding performance, excelling in FID and LPIPS with 0.01 and 36.99 ahead of the second place in the private set. In the public dataset, DiffuCE has a competitive performance compared to other SOTAs. Moreover, in expert assessments, DiffuCE achieves the highest score of 7.06 for overall satisfaction, which is 1.38 ahead of the second place, affirming its performance from a clinical stand-point. Codes are available at https://github.com/lzh107u/DiffuCE

Recommendation systems have become more important since the widespread use of the Internet. The large amount of information means that it is difficult for users to discover what they really need. However, users’ preferences can change over time because of the age and the impact of social networks. Recommendation systems must capture users’ preferences and recommend suitable items, which is a difficult challenge. This study proposes a novel recommendation system that adapts to the changing preferences of users. By learning and adapting to users’ changing preferences better recommendations are provided. This study uses context factors as additional information to model users’ preferences more accurately. For the proposed recommendation system, the context is based on users’ most recent interaction items. This study proposes two novel attention mechanisms for the recommendation system. The contextual item attention module captures contextual information, changing pattern in users’ preference and the importance of items. The multi-head attention module extends the diversity of users’ preferences and adapts to changing preferences. The recommendation performance is improved using additional item's temporal information to model the contextual item's representation. Experiments compare the proposed algorithm with several state-of-the-art recommendation methods using three real-world datasets. The experimental results demonstrate that the proposed context-aware recommendation model outperforms traditional methods and demonstrate the effectiveness with which contextual information is captured by the attention mechanism.

Mutational signatures help identify cancer-associated genes that are being involved in tumorigenesis pathways. Hence, these pathways guide precision medicine approaches to find appropriate drugs and treatments. The pattern of mutations varies in different cancer types. Some mutations dysregulate protein function so that their accumulation is responsible for cancer development and might be associated with different cancer types. Therefore, mutations as a feature set can be used as an informative candidate to distinguish various cancer types. There are several options for demonstrating mutations. One might employ binary values to demonstrate mutation regions. Another potential method for extracting features is utilizing mutation interpreters. In this study, we investigate the trinucleotide mutational pattern of each cancer type. Moreover, we extract salient NMF-based mutational signatures across various cancer types. Then, we identify cancer-associated genes of a target cancer based on its salient signatures. We evaluate the cancer-associated genes using survival and gene expression analysis in different stages of cancer. Furthermore, we introduce DiaDeL, which is a deep learning-based binary classifier. The DiaDeL model uses mutational signatures as input features and distinct a cancer type from the others. Our proposed model outperforms six state-of-the-art methods with 0.824 and 0.88 for accuracy and AUC, respectively. The source code is available at https://github.com/sabdollahi/DiaDeL.

Hypotheses are the most important part of medical research. If we have a good hypothesis, we can design experiments and verify it. Therefore, we use the associations generated by association rule as hypotheses in clinical medicine research. We hope this method can help physicians quickly and correctly find research hypotheses. This experiment was divided into two parts. In the first part, we used the Apriori algorithm to find associations between cancer and other catastrophic illnesses. In the second part, we used these associations as medical-research hypotheses and designed cohort studies to verify them. In this study, we proved that the association-rules method could help clinical physicians quickly and correctly obtain clinical-medicine hypotheses.

Several studies to date have proposed different types of interpreters for measuring the degree of pathogenicity of variants. However, in predicting the disease type and disease-gene associations, scholars face two essential challenges, namely the vast number of existing variants and the existence of variants which are recognized as variant of uncertain significance (VUS). To tackle these challenges, we propose algorithms to assign a significance to each gene rather than each variant, describing its degree of pathogenicity. Since the interpreters identified most of the variants as VUS, most of the gene scores were identified as uncertain significance. To predict the uncertain significance scores, we design two matrix factorization-based models: the common latent space model uses genomics variant data as well as heterogeneous clinical data, while the single-matrix factorization model can be used when heterogeneous clinical data are unavailable. We have managed to show that the models successfully predict the uncertain significance scores with low error and high accuracy. Moreover, to evaluate the effectiveness of our novel input features, we train five different multi-label classifiers including a feedforward neural network with the same feature set and show they all achieve high accuracy as the main impact of our approach comes from the features. Availability: The source code is freely available at https://github.com/sabdollahi/CoLaSpSMFM.

Biophysical protein-protein interactions perform dominant roles in the initiation and progression of many cancer-related pathways. A protein-protein interaction might play different roles in diverse cancer types. Hence, prioritizing the PPIs in each cancer type would help detect cancer-associated pathways, find a better understanding of cancer biology, and facilitate drug discovery. Several studies to date have proposed computational methods for extracting the PPI essentiality of different cancer types based on the PPI network. The main drawback of these studies is not using a rich source such as genomics variant data. An amino acid sequence encodes useful information about protein structure and behavior. We represent each amino acid sequence based on its variants/mutations in seven different ways: binary vectors, pathogenicity scores, binding affinity changes upon mutations, gene expression-based network of the interactions, biophysicochemical properties, g-gap dipeptide, and one-hot vectors. Based on these representations, we design and consider seven different deep learning models. Then, we compare the accuracy of these models in predicting 20 different cancer types from the TCGA cohort. WinBinVec is a window-based model that outperforms the other models. Moreover, WinBinVec contains a PPI essentiality module that helps extract the essentiality probability of each PPI for every cancer type. Source code and Data: https://github.com/sabdollahi/WinBinVec.

mender system is studied widely and has been implemented successfully into businesses and daily lives. Its primary purpose is to analyze user behaviors to predict the next item of interest. To enrich user information, a session-based recommender system considers the user's historical records in the session. This information includes the user's general and current interests as additional information to improve the performance of the model such that the next item can be recommended easily. However, the recommender system not only considers the prediction performance but also regards the interpretability as a major target. For enhancing the interpretability and performance, we propose a novel mechanism, i.e., FuzzAttention, based on the attention mechanism that is applied widely to deep-learning models. In FuzzAttention, we utilize a fuzzy neural network to build a fuzzy inference system; therefore, we adopt joint learning to learn the parameters in the session-based recommendation and the fuzzy neural network jointly. In the experiments, we used two types of session-based recommender systems and conducted them on two datasets including the session-based information to compare the model performance based on the traditional attention mechanism and FuzzAttention. The results indicate that our proposed mechanism can improve the performance to predict the next item and the model's interpretability.

Drug development is an expensive and time-consuming process; these could be reduced if the existing resources could be used to identify candidates for drug repurposing. This study sought to do this by text mining a large-scale literature repository to curate repurposed drug lists for different cancers. We devised a pattern-based relationship extraction method to extract disease-gene and gene-drug direct relationships from the literature. These direct relationships are used to infer indirect relationships using the ABC model. A gene-shared ranking method based on drug target similarity was then proposed to prioritize the indirect relationships. Our method of assessing drug target similarity correlated to existing anatomical therapeutic chemical code-based methods with a Pearson correlation coefficient of 0.9311. The indirect relationships ranking method achieved a significant mean average precision score of top 100 most common diseases. We also confirmed the suitability of candidates identified for repurposing as anticancer drugs by conducting a manual review of the literature and the clinical trials. Eventually, for visualization and enrichment of huge amount of repurposed drug information, a chord diagram was demonstrated to rapidly identify two novel indications for further biological evaluations.

MOTIVATION: With the spreading technique of mass sequencing, nucleosome positions and scores for their intensity have become available through several previous studies in yeast, but relatively few studies have specifically aimed to determine the score of nucleosome stability. Based on mass sequencing data, we proposed a nucleosome center score (NCS) for quantifying nucleosome stability by measuring shifts of the nucleosome center, and then mapping NCS scores to nucleosome positions in Brogaard et al.'s study. RESULTS: We demonstrated the efficiency of NCS by known preference of A/T-based tracts for nucleosome formation, and showed that central nucleosomal DNA is more sensitive to A/T-based tracts than outer regions, which corresponds to the central histone tetramer-dominated region. We also found significant flanking preference around nucleosomal DNA for A/T-based dinucleotides, suggesting that neighboring sequences could affect nucleosome stability. Finally, the difference between results of NCS and Brogaard et al.'s scores was addressed and discussed. CONTACTS: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Protein phosphorylation is critical for regulating cellular activities by controlling protein activities, localization and turnover, and by transmitting information within cells through signaling networks. However, predictions of protein phosphorylation and signaling networks remain a significant challenge, lagging behind predictions of transcriptional regulatory networks into which they often feed. RESULTS: We developed PhosphoChain to predict kinases, phosphatases and chains of phosphorylation events in signaling networks by combining mRNA expression levels of regulators and targets with a motif detection algorithm and optional prior information. PhosphoChain correctly reconstructed ∼78% of the yeast mitogen-activated protein kinase pathway from publicly available data. When tested on yeast phosphoproteomic data from large-scale mass spectrometry experiments, PhosphoChain correctly identified ∼27% more phosphorylation sites than existing motif detection tools (NetPhosYeast and GPS2.0), and predictions of kinase-phosphatase interactions overlapped with ∼59% of known interactions present in yeast databases. PhosphoChain provides a valuable framework for predicting condition-specific phosphorylation events from high-throughput data. AVAILABILITY: PhosphoChain is implemented in Java and available at http://virgo.csie.ncku.edu.tw/PhosphoChain/ or http://aitchisonlab.com/PhosphoChain

MOTIVATION: Determination of the binding affinity of a protein-ligand complex is important to quantitatively specify whether a particular small molecule will bind to the target protein. Besides, collection of comprehensive datasets for protein-ligand complexes and their corresponding binding affinities is crucial in developing accurate scoring functions for the prediction of the binding affinities of previously unknown protein-ligand complexes. In the past decades, several databases of protein-ligand-binding affinities have been created via visual extraction from literature. However, such approaches are time-consuming and most of these databases are updated only a few times per year. Hence, there is an immediate demand for an automatic extraction method with high precision for binding affinity collection. RESULT: We have created a new database of protein-ligand-binding affinity data, AutoBind, based on automatic information retrieval. We first compiled a collection of 1586 articles where the binding affinities have been marked manually. Based on this annotated collection, we designed four sentence patterns that are used to scan full-text articles as well as a scoring function to rank the sentences that match our patterns. The proposed sentence patterns can effectively identify the binding affinities in full-text articles. Our assessment shows that AutoBind achieved 84.22% precision and 79.07% recall on the testing corpus. Currently, 13 616 protein-ligand complexes and the corresponding binding affinities have been deposited in AutoBind from 17 221 articles. AVAILABILITY: AutoBind is automatically updated on a monthly basis, and it is freely available at http://autobind.csie.ncku.edu.tw/ and http://autobind.mc.ntu.edu.tw/. All of the deposited binding affinities have been refined and approved manually before being released.

MOTIVATION: The growing availability of full-text scientific articles raises the important issue of how to most efficiently digest full-text content. Although article titles and abstracts provide accurate and concise information on an article's contents, their brevity inevitably entails the loss of detail. Full-text articles provide those details, but require more time to read. The primary goal of this study is to combine the advantages of concise abstracts and detail-rich full-texts to ease the burden of reading. RESULTS: We retrieved abstract-related paragraphs from full-text articles through shared keywords between the abstract and paragraphs from the main text. Significant paragraphs were then recommended by applying a proposed paragraph ranking approach. Finally, the user was provided with a condensed text consisting of these significant paragraphs, allowing the user to save time from perusing the whole article. We compared the performance of the proposed approach with a keyword counting approach and a PageRank-like approach. Evaluation was conducted in two aspects: the importance of each retrieved paragraph and the information coverage of a set of retrieved paragraphs. In both evaluations, the proposed approach outperformed the other approaches. CONTACT: [email protected].

The objective of this study was to develop and validate an automated acquisition system to assess quality of care (QC) measures for cardiovascular diseases. This system combining searching and retrieval algorithms was designed to extract QC measures from electronic discharge notes and to estimate the attainment rates to the current standards of care. It was developed on the patients with ST-segment elevation myocardial infarction and tested on the patients with unstable angina/non-ST-segment elevation myocardial infarction, both diseases sharing almost the same QC measures. The system was able to reach a reasonable agreement (kappa value) with medical experts from 0.65 (early reperfusion rate) to 0.97 (beta-blockers and lipid-lowering agents before discharge) for different QC measures in the test set, and then applied to evaluate QC in the patients who underwent coronary artery bypass grafting surgery. The result has validated a new tool to reliably extract QC measures for cardiovascular diseases.

When analyzing the results of microarray experiments, biologists generally use unsupervised categorization tools. However, such tools regard each time point as an independent dimension and utilize the Euclidean distance to compute the similarities between expressions. Furthermore, some of these methods require the number of clusters to be determined in advance, which is clearly impossible in the case of a new dataset. Therefore, this study proposes a novel scheme, designated as the Variation-based Coexpression Detection (VCD) algorithm, to analyze the trends of expressions based on their variation over time. The proposed algorithm has two advantages. First, it is unnecessary to determine the number of clusters in advance since the algorithm automatically detects those genes whose profiles are grouped together and creates patterns for these groups. Second, the algorithm features a new measurement criterion for calculating the degree of change of the expressions between adjacent time points and evaluating their trend similarities. Three real-world microarray datasets are employed to evaluate the performance of the proposed algorithm.

Proteomics technologies and bioinformatics tools have been widely used to analyze protein-protein interactions of complex biological systems, which are essential for understanding the mechanisms of human and cancer biology. Although many studies have tackled the problem of high-throughput protein-protein interaction identifications inSaccharomycescerevisiae,Caenorhabditiselegans, andDrosophilamelanogaster, the effort to predict human and cancer-related protein-protein interaction is still limited. Moreover, low consistency and high false positive rates are major drawbacks of these high-throughput methods. In this research, the focus is on predicting human cancer-related protein-protein interaction and reducing false positive rates with integrated classifiers. We propose a hybrid machine learning system by merging fuzzy multiset-based classifiers and support vector machines (SVMs) into fuzzy-SVM mixture models (FSMMs). Our experimental result of the FSMMs approach achieves consistent prediction accuracy on human protein-protein interactions (PPIs) with an receiver operating curve score of 0.965 that outperforms other models. Overall, prediction results on cancer-related protein pairs indicate that our proposed system is effective for identifying both known and novel PPIs to assist cancer research in discovering novel interactions.

Support vector machines (SVMs) have been very successful in pattern classification and function approximation problems for crisp data. In this paper, we incorporate the concept of fuzzy set theory into the support vector regression machine. The parameters to be estimated in the SVM regression, such as the components within the weight vector and the bias term, are set to be the fuzzy numbers. This integration preserves the benefits of SVM regression model and fuzzy regression model and has been attempted to treat fuzzy nonlinear regression analysis. In contrast to previous fuzzy nonlinear regression models, the proposed algorithm is a model-free method in the sense that we do not have to assume the underlying model function. By using different kernel functions, we can construct different learning machines with arbitrary types of nonlinear regression functions. Moreover, the proposed method can achieve automatic accuracy control in the fuzzy regression analysis task. The upper bound on number of errors is controlled by the user-predefined parameters. Experimental results are then presented that indicate the performance of the proposed approach.

This paper proposes a new clustering algorithm referred to as the possibilitic latent variables (PLV) clustering algorithm. This algorithm provides a powerful tool for the analysis of complex data, such as clinical diagnosis and biological expressions data, due to its robustness to various data distributions and its accuracy in establishing appropriate groups from data. The algorithm combines a distribution model and the fuzzy degrees concept. Compared to the expectation-maximization (EM) algorithm, which is a well-known distribution estimating algorithm, the PLV algorithm has the considerable advantage that it can be applied to various data types, i.e. it is not restricted solely to Gaussian data distributions. Additionally, the proposed algorithm has a better performance than the well-known fuzzy clustering algorithm, i.e. the FCM algorithm, where it can address compact regions, other than simply dividing objects into several equal populations. The performance of the proposed algorithm is verified by conducting clustering tasks on the contents of several medical diagnosis and biological expressions datasets.

BACKGROUND: Modeling cancer-related regulatory modules from gene expression profiling of cancer tissues is expected to contribute to our understanding of cancer biology as well as developments of new diagnose and therapies. Several mathematical models have been used to explore the phenomena of transcriptional regulatory mechanisms in Saccharomyces cerevisiae. However, the contemplating on controlling of feed-forward and feedback loops in transcriptional regulatory mechanisms is not resolved adequately in Saccharomyces cerevisiae, nor is in human cancer cells. RESULTS: In this study, we introduce a Genetic Algorithm-Recurrent Neural Network (GA-RNN) hybrid method for finding feed-forward regulated genes when given some transcription factors to construct cancer-related regulatory modules in human cancer microarray data. This hybrid approach focuses on the construction of various kinds of regulatory modules, that is, Recurrent Neural Network has the capability of controlling feed-forward and feedback loops in regulatory modules and Genetic Algorithms provide the ability of global searching of common regulated genes. This approach unravels new feed-forward connections in regulatory models by modified multi-layer RNN architectures. We also validate our approach by demonstrating that the connections in our cancer-related regulatory modules have been most identified and verified by previously-published biological documents. CONCLUSION: The major contribution provided by this approach is regarding the chain influences upon a set of genes sequentially. In addition, this inverse modeling correctly identifies known oncogenes and their interaction genes in a purely data-driven way.

BACKGROUND: Abundant information about gene products is stored in online searchable databases such as annotation or literature. To efficiently obtain and digest such information, there is a pressing need for automated information-summarization and functional-similarity clustering of genes. RESULTS: We have developed a novel method for semantic measurement of annotation and integrated it with a biomedical literature summarization system to establish a platform, GeneLibrarian, to provide users well-organized information about any specific group of genes (e.g. one cluster of genes from a microarray chip) they might be interested in. The GeneLibrarian generates a summarized viewgraph of candidate genes for a user based on his/her preference and delivers the desired background information effectively to the user. The summarization technique involves optimizing the text mining algorithm and Gene Ontology-based clustering method to enable the discovery of gene relations. CONCLUSION: GeneLibrarian is a Java-based web application that automates the process of retrieving critical information from the literature and expanding the number of potential genes for further analysis. This study concentrates on providing well organized information to users and we believe that will be useful in their researches. GeneLibrarian is available on http://gen.csie.ncku.edu.tw/GeneLibrarian/.

This paper presents the pruning and model-selecting algorithms to the support vector learning for sample classification and function regression. When constructing RBF network by support vector learning we occasionally obtain redundant support vectors which do not significantly affect the final classification and function approximation results. The pruning algorithms primarily based on the sensitivity measure and the penalty term. The kernel function parameters and the position of each support vector are updated in order to have minimal increase in error, and this makes the structure of SVM network more flexible. We illustrate this approach with synthetic data simulation and face detection problem in order to demonstrate the pruning effectiveness.

With the rapid growth of articles of genomics research, it has become a challenge for biomedical researchers to access this ever-increasing quantity of information to understand the newest discovery of functions of proteins they are studying. To facilitate functional annotation of proteins by utilizing the huge amounts of biomedical literature and transforming the knowledge into easily accessible database formats, the text mining technique thus becomes essential. In this paper, we propose the method of sentence pattern mining to extract protein functions from biomedical literature. To recognize variants of function terms correctly, we identify morphological, syntactic, and semantic variation forms. The proposed methods can be used to aid database curators in annotating protein functions and to assist biologists and medical researchers in searching protein functions from biomedical literature.

We have developed a gene-gene (G-G) relation browser that combines fuzzy sequential pattern mining and information-extraction model to extract from biomedical literature knowledge on gene-gene interactions. Our approach aims to detect associated G-G relations that are often discussed in documents. Integration of the related relations lead to an individual G-G network. Graphic presentations are to be used to demonstrate the relationships between gene products.

UNLABELLED: We present a biomedical text-mining system focused on four types of gene-related information: biological functions, associated diseases, related genes and gene-gene relations. The aim of this system is to provide researchers an easy-to-use bio-information service that will rapidly survey the rapidly burgeoning biomedical literature. AVAILABILITY: http://iir.csie.ncku.edu.tw/~yuhc/gis/

This article presents an intelligent news recommender agent (INRA), which can be used to filter news articles as well as to recommend relevant news for individual user automatically. Three specific objectives underlie the presentation of the intelligent news recommender agent in this study. The first is to describe the basic architecture of this approach, and the second is to show the design of the fuzzy hierarchical mixture of the expert model for text categorization. The third and more elaborate goal is to show that the proposed system is able to perform a news-recommending process. We show this approach with standard benchmark examples of the Reuters-21578 in order to verify the effectiveness of news recommending. © 2004 Wiley Periodicals, Inc.

This paper describes a fuzzy modeling framework based on support vector machine, a rule-based framework that explicitly characterizes the representation in fuzzy inference procedure. The support vector learning mechanism provides an architecture to extract support vectors for generating fuzzy IF-THEN rules from the training data set, and a method to describe the fuzzy system in terms of kernel functions. Thus, it has the inherent advantage that the model does not have to determine the number of rules in advance, and the overall fuzzy inference system can be represented as series expansion of fuzzy basis functions. The performance of the proposed approach is compared to other fuzzy rule-based modeling methods using four data sets.

This paper presents a new fuzzy cover-based clustering algorithm. In the proposed algorithm, the concept of fuzzy cover and objective function are employed to identify holding points in the dataset, and we associate these holding points together to build up the backbones of the final clusters. Three specific objectives underlie the presentation of the proposed approach in this paper. The first is to describe mathematical formulation of the fuzzy covers, and the second is to summarize the detailed procedure of constructing fuzzy covers and splicing them into clusters. The third goal is to demonstrate that this approach is able to find out reasonable representative patterns in the final clusters. We illustrate this approach with four examples in order to verify the clustering effectiveness.

MOTIVATION: Research on roles of gene products in cells is accumulating and changing rapidly, but most of the results are still reported in text form and are not directly accessible by computers. To expedite the progress of functional bioinformatics, it is, therefore, important to efficiently process large amounts of biomedical literature and transform the knowledge extracted into a structured format usable by biologists and medical researchers. Our aim was to develop an intelligent text-mining system that will extract from biomedical documents knowledge about the functions of gene products and thus facilitate computing with function. RESULTS: We have developed an ontology-based text-mining system to efficiently extract from biomedical literature knowledge about the functions of gene products. We also propose methods of sentence alignment and sentence classification to discover the functions of gene products discussed in digital texts. AVAILABILITY: http://ismp.csie.ncku.edu.tw/~yuhc/meke/

In this paper, the support vector clustering is extended to an adaptive cell growing model which maps data points to a high dimensional feature space through a desired kernel function. This generalized model is called multiple spheres support vector clustering, which essentially identifies dense regions in the original space by finding their corresponding spheres with minimal radius in the feature space. A multisphere clustering algorithm based on adaptive cluster cell growing method is developed, whereby it is possible to obtain the grade of memberships, as well as cluster prototypes in partition. The effectiveness of the proposed algorithm is demonstrated for the problem of arbitrary cluster shapes and for prototype identification in an actual application to a handwritten digit data set.

In this paper, we present a document categorization method based on the hierarchical fuzzy networks. The proposed model employs the divide-and-conquer principle to resolve documents categorization problem based on a predefined hierarchical structure. The final classification framework can be interpreted as a hierarchical array of non-linear decision tree. Each node in the tree represents one filter. The fuzzy K-nearest-neighbor (KNN)-based filter decides that the unknown document belongs to the corresponding category or not. We use the Reuters-21578 news data set to evaluate the performance of the proposed method.

Recent advances in network-based decision making methods have given rise to computationally efficient solution methodologies for intelligent systems. One type of hierarchical network implementation, the fuzzy integral operator approach, is investigated. In this approach, we generalized the Choquet fuzzy integral as an excellent component for decision analysis and making. This involves extending the standard operators in information aggregation with generalized operators, resulting in increased flexibility. The characteristics of the Choquet fuzzy integrals and their generalizations are addressed and network-based decision making frameworks are then proposed. The trainable hierarchical networks are able to perceive and interpret complex decisions by using those processing elements called neurons. We also present a decision making experiment using the proposed network to learn appropriate functional relationships in the defective numeric fields detection domain. ©1999 John Wiley & Sons, Inc.

A Choquet fuzzy integral-based approach to hierarchical network implementation is investigated. In this approach, we generalized the fuzzy integral as an excellent component for decision analysis. The generalization involves replacing the max (or min) operator in information aggregation with a fuzzy integral-based neuron, resulting in increased flexibility. The characteristics of the Choquet fuzzy integral are studied and a network-based decision-analysis framework is proposed. The trainable hierarchical network can be implemented utilizing the fuzzy integral-based neurons and connectives. The training algorithms are derived and several examples given to illustrate the behaviors of the networks. Also, we present a decision making experiment using the proposed network to learn appropriate functional relationships in the defective numeric fields detection domain.

We remark on an error in the above paper by Chiang (ibid. vol.7 (1999)). The purpose of this note is to present the correct formulas for partial derivatives of fuzzy integral-based neural nodes with respect to densities of Sugeno measures. In reply, Chiang agree with the correction.

Two hybrid fuzzy neural systems are developed and applied to handwritten word recognition. The word recognition system requires a module that assigns character class membership values to segments of images of handwritten words. The module must accurately represent ambiguities between character classes and assign low membership values to a wide variety of noncharacter segments resulting from erroneous segmentations. Each hybrid is a cascaded system. The first stage of both is a self-organizing feature map (SOFM). The second stages map distances into membership values. The third stage of one system is a multilayer perceptron (MLP). The third stage of the other is a bank of Choquet fuzzy integrals (FI). The two systems are compared individually and as a combination to the baseline system. The new systems each perform better than the baseline system. The MLP system slightly outperforms the FI system, but the combination of the two outperforms the individual systems with a small increase in computational cost over the MLP system. Recognition rates of over 92% are achieved with a lexicon set having average size of 100. Experiments were performed on a standard test set from the SUNY/USPS CD-ROM database.

An off-line handwritten word recognition system is described. Images of handwritten words are matched to lexicons of candidate strings. A word image is segmented into primitives. The best match between sequences of unions of primitives and a lexicon string is found using dynamic programming. Neural networks assign match scores between characters and segments. Two particularly unique features are that neural networks assign confidence that pairs of segments are compatible with character confidence assignments and that this confidence is integrated into the dynamic programming. Experimental results are provided on data from the U.S. Postal Service.

A hybrid feature extraction framework for handwritten numeric fields recognition is described. The numeric fields were extracted from binary images of credit card application forms. The images include identity numbers (ID) and phone numbers. The feature extraction framework utilizes a cascade of a Kohonen self-organizing feature map (SOM) and a set of non-linear filtering units. The goals of our feature extraction process are to provide reliable information to the recognition stage. The recognition stage uses the feature set as inputs to a multi-layer neural network. We present experimental results which demonstrate the ability to extract features automatically in handwritten digit recognition. Experiments were performed on a test data set from the CCL/ITRI Database which consists of over 90,390 handwritten numeric digits. Recognition rate of 98.6% is achieved on this database.

Experiments comparing neural networks trained with crisp and fuzzy desired outputs are described. A handwritten word recognition algorithm using the neural networks for character level confidence assignment was tested on images of words taken from the United States Postal Service mailstream. The fuzzy outputs were defined using a fuzzy k-nearest neighbor algorithm. The crisp networks slightly outperformed the fuzzy networks at the character level but the fuzzy networks outperformed the crisp networks at the word level. This empirical result is interpreted as an example of the principle of least commitment.>