Stock classification is a challenging task due to high levels of noise and volatility of stocks returns. In this paper we show that using transfer learning can help with this task, by pre-training a model to extract universal features on the full universe of stocks of the S&P500 index and then transferring it to another model to directly learn a trading rule. Transferred models present more than double the risk-adjusted returns than their counterparts trained from zero. In addition, we propose the use of data augmentation on the feature space defined as the output of a pre-trained model (i.e. augmenting the aggregated time-series representation). We compare this augmentation approach with the standard one, i.e. augmenting the time-series in the input space. We show that augmentation methods on the feature space leads to 20% increase in risk-adjusted return compared to a model trained with transfer learning but without augmentation.

The process of preparing potentially large and complex data sets for further analysis or manual examination is often called data wrangling. In classical warehousing environments, the steps in such a process are carried out using Extract-Transform-Load platforms, with significant manual involvement in specifying, configuring or tuning many of them. In typical big data applications, we need to ensure that all wrangling steps, including web extraction, selection, integration and cleaning, benefit from automation wherever possible. Towards this goal, in the paper we: (i) introduce a notion of data context, which associates portions of a target schema with extensional data of types that are commonly available; (ii) define a scalable methodology to bootstrap an end-to-end data wrangling process based on data profiling; (iii) describe how data context is used to inform automation in several steps within wrangling, specifically, matching, value format transformation, data repair, and mapping generation and selection to optimise the accuracy, consistency and relevance of the result; and (iv) we evaluate the approach with real estate data and financial data, showing substantial improvements in the results of automated wrangling.

We report experience in requirements elicitation of domain knowledge from experts in clinical and cognitive neurosciences. The elicitation target was a causal model for early signs of dementia indicated by changes in user behaviour and errors apparent in logs of computer activity. A Delphi-style process consisting of workshops with experts followed by a questionnaire was adopted. The paper describes how the elicitation process had to be adapted to deal with problems encountered in terminology and limited consensus among the experts. In spite of the difficulties encountered, a partial causal model of user behavioural pathologies and errors was elicited. This informed requirements for configuring data- and text-mining tools to search for the specific data patterns. Lessons learned for elicitation from experts are presented, and the implications for requirements are discussed as “unknown unknowns”, as well as configuration requirements for directing data-/text-mining tools towards refining awareness requirements in healthcare applications.

Traditionally, predictive maintenance of wind turbines has relied on experts to perform time consuming feature pre-processing using statistical, time and frequency domain analysis. Recent advancements in Convolutional Neural Networks have opened the potential for using featureless approaches that learn the discriminating patterns from big data sets without expert intervention. Given multi-dimensional time series data representing sensed electric currents, in this paper we explore the optimal window length that can be used to accurately determine its speed and load using Convolutional and Residual Networks. Choosing an optimal window length is a trade-off between accuracy and number of predictions per second. Fast predictions for operating parameters are useful in maintenance strategies but they come at a cost of decreased accuracy (as there is less data available in a shorter time interval). We show how fusing multiple signals can achieve high accuracy with nine speed and load cases varying from 375rpm with 0% load to 1500rpm and 100% load. Using Class Activation Maps we investigate features in the time domain picked up by the network to make its classification decisions. Further, we train the networks to predict the drive loads in a regression setting where the models are able to generalize well on unseen cases.

Data scientists are usually interested in a subset of sources with properties that are most aligned to intended data use. The SOURCERY system supports interactive multi-criteria user-driven source selection. SOURCERY allows a user to identify criteria they consider of importance and indicate their relative importance, and seeks a source selection result aligned to the user-supplied criteria preferences. The user is given an overview of the properties of the sources that are selected along with visual analyses contextualizing the result in relation to what is theoretically possible and what is possible given the set of available sources. The system also enables a user to interactively perform iterative fine-tuning to explore how changes to preferences may impact results.

Standard neural network architectures are non-linear only by virtue of a simple element-wise activation function, making them both brittle and excessively large. In this paper, we consider methods for making the feed-forward layer more flexible while preserving its basic structure. We develop simple drop-in replacements that learn to adapt their parameterization conditional on the input, thereby increasing statistical efficiency significantly. We present an adaptive LSTM that advances the state of the art for the Penn Treebank and Wikitext-2 word-modeling tasks while using fewer parameters and converging in half as many iterations.

Source selection is the problem of identifying a subset of available data sources that best meet a user’s needs. In this paper we propose a user-driven approach to source selection that seeks to identify sources that are most fit for purpose. The approach employs a decision support methodology to take account of a user’s context, to allow end users to tune their preferences by specifying the relative importance between different criteria, looking to find a trade-off solution aligned with his/her preferences. The approach is extensible to incorporate diverse criteria, not drawn from a fixed set, and solutions can use a subset of the data from each selected source, rather than require that sources are used in their entirety or not at all. The paper describes and motivates the approach, presenting a methodology for modelling a user’s context, and its collection of optimisation algorithms for exploring the space of solutions, and compares and evaluates the resulting algorithms using multiple real world data sets. The experiments show how source selection results are produced that are attuned to each user’s preferences, both with respect to overall weighted utility and through faithful representation of a user’s preferences within a result, while scaling to potentially thousands of sources.

The process of preparing potentially large and complex data sets for further analysis or manual examination is often called data wrangling. In classical warehousing environments, the steps in such a process have been carried out using Extract-Transform-Load platforms, with significant manual involvement in specifying, configuring or tuning many of them. Cost-effective data wrangling processes need to ensure that data wrangling steps benefit from automation wherever possible. In this paper, we define a methodology to fully automate an end-to-end data wrangling process incorporating data context, which associates portions of a target schema with potentially spurious extensional data of types that are commonly available. Instance-based evidence together with data profiling paves the way to inform automation in several steps within the wrangling process, specifically, matching, mapping validation, value format transformation, and data repair. The approach is evaluated with real estate data showing substantial improvements in the results of automated wrangling.

Data wrangling, the multi-faceted process by which the data required by an application is identified, extracted, cleaned and integrated, is often cumbersome and labor intensive. In this paper, we present an architecture that supports a complete data wrangling lifecycle, orchestrates components dynamically, builds on automation wherever possible, is informed by whatever data is available, refines automatically produced results in the light of feedback, takes into account the user's priorities, and supports data scientists with diverse skill sets. The architecture is demonstrated in practice for wrangling property sales and open government data.

Fuzzy C-means has been utilized successfully in a wide range of applications, extending the clustering capability of the K-means to datasets that are uncertain, vague and otherwise hard to cluster. This paper introduces the Fuzzy C-means++ algorithm which, by utilizing the seeding mechanism of the K-means++ algorithm, improves the effectiveness and speed of Fuzzy C-means. By careful seeding that disperses the initial cluster centers through the data space, the resulting Fuzzy C-means++ approach samples starting cluster representatives during the initialization phase. The cluster representatives are well spread in the input space, resulting in both faster convergence times and higher quality solutions. Implementations in R of standard Fuzzy C-means and Fuzzy C-means++ are evaluated on various data sets. We investigate the cluster quality and iteration count as we vary the spreading factor on a series of synthetic data sets. We run the algorithm on real world data sets and to account for the non-determinism inherent in these algorithms we record multiple runs while choosing different k parameter values. The results show that the proposed method gives significant improvement in convergence times (the number of iterations) of up to 40 (2.1 on average) times the standard on synthetic datasets and, in general, an associated lower cost function value and Xie–Beni value. A proof sketch of the logarithmically bounded expected cost function value is given.

In group decision making, multiple decision makers (DMs) aim to reach a consensus ranking of alternatives in a decision problem. The differing expertise, experience and, potentially conflicting, interests of the DMs will result in the need for some form of conciliation to achieve consensus. Pairwise comparisons are commonly used to elicit values of preference of a DM. The aggregation of the preferences of multiple DMs must additionally consider potential conflict between DMs and how these conflicts may result in a need for compromise to reach group consensus. We present an approach to aggregating the preferences of multiple DMs, utilizing multi-objective optimization, to derive and highlight underlying conflict between the DMs when seeking to achieve consensus. Extracting knowledge of conflict facilitates both traceability and transparency of the trade-offs involved when reaching a group consensus. Further, the approach incorporates inconsistency reduction during the aggregation process to seek to diminish adverse effects upon decision outcomes. The approach can determine a single final solution based on either global compromise information or through utilizing weights of importance of the DMs. Within multi-criteria decision making, we present a case study within the Analytical Hierarchy Process from which we derive a richer final ranking of the decision alternatives.

A recent promise to access unstructured clinical data from electronic health records on large-scale has revitalized the interest in automated de-identification of clinical notes, which includes the identification of mentions of Protected Health Information (PHI). We describe the methods developed and evaluated as part of the i2b2/UTHealth 2014 challenge to identify PHI defined by 25 entity types in longitudinal clinical narratives. Our approach combines knowledge-driven (dictionaries and rules) and data-driven (machine learning) methods with a large range of features to address de-identification of specific named entities. In addition, we have devised a two-pass recognition approach that creates a patient-specific run-time dictionary from the PHI entities identified in the first step with high confidence, which is then used in the second pass to identify mentions that lack specific clues. The proposed method achieved the overall micro F1-measures of 91% on strict and 95% on token-level evaluation on the test dataset (514 narratives). Whilst most PHI entities can be reliably identified, particularly challenging were mentions of Organizations and Professions. Still, the overall results suggest that automated text mining methods can be used to reliably process clinical notes to identify personal information and thus providing a crucial step in large-scale de-identification of unstructured data for further clinical and epidemiological studies.

Heart disease is the leading cause of death globally and a significant part of the human population lives with it. A number of risk factors have been recognized as contributing to the disease, including obesity, coronary artery disease (CAD), hypertension, hyperlipidemia, diabetes, smoking, and family history of premature CAD. This paper describes and evaluates a methodology to extract mentions of such risk factors from diabetic clinical notes, which was a task of the i2b2/UTHealth 2014 Challenge in Natural Language Processing for Clinical Data. The methodology is knowledge-driven and the system implements local lexicalized rules (based on syntactical patterns observed in notes) combined with manually constructed dictionaries that characterize the domain. A part of the task was also to detect the time interval in which the risk factors were present in a patient. The system was applied to an evaluation set of 514 unseen notes and achieved a micro-average F-score of 88% (with 86% precision and 90% recall). While the identification of CAD family history, medication and some of the related disease factors (e.g. hypertension, diabetes, hyperlipidemia) showed quite good results, the identification of CAD-specific indicators proved to be more challenging (F-score of 74%). Overall, the results are encouraging and suggested that automated text mining methods can be used to process clinical notes to identify risk factors and monitor progression of heart disease on a large-scale, providing necessary data for clinical and epidemiological studies.

Biomedical and genomic data are inherently heterogeneous and their recent proliferation over the Web has demanded innovative querying methods to help domain experts in their clinical and research studies. In this paper we present the use of Semantic Web technologies in querying diverse phenotype-genotype associations for supporting personalized medicine and potentially helping to discover new associations. Our initial results suggest that Semantic Web technologies has competitive advantages in extracting, consolidating and presenting phenotype-genotype associations that resides in various bioinformatics resources. The developed querying method could support researchers and medical professionals in discovering and utilizing information on published associations relating disease, treatment, adverse events and environmental factors to genetic markers from multiple repositories.

Pair wise comparisons are commonly used to estimate values of preference among a finite set of decision alternatives with regards to intangible factors. Inconsistency within decision making judgments may occur. This work proposes an approach to reducing inconsistency using multi-objective optimization with the objectives of different inconsistency types and judgment modification measures. The approach allows the decision maker to choose both the inconsistency measure(s) and the modification measure(s) employed to suit their needs and attitudes. Utilizing multi-objective optimization allows for a range of possible tradeoff solutions to be presented to the decision maker for selection, aiding them in their pursuit of inconsistency reduction. It also enables better understanding of the characteristics of the decision problem and its inconsistency.

In addition to structured data, electronic health records contain unstructured clinical notes and narratives. The identification and classification of mentions of relevant clinical concepts is a crucial preprocessing step in designing and developing clinical decision support systems. While this task has gained significant attention in recent years, there are still a number of issues that need further investigation. This paper explores a variety of common challenges faced by clinical named entity recognition and classification methods as well as current approaches to handling them.

Low back pain (LBP) affects most people at some time in their life and psychological factors are often viewed as obstacles to recovery. LBP is often accompanied by hyperactivity of superficial Para spinal muscles and it has been suggested that psychological factors may affect the condition via increased spinal loading resulting from altered Para spinal muscle activity. Several measurements are taken, including physical factors (muscle activity, pain intensity, disability) and psychosocial factors (anxiety, depression, fear of movement etc) using several numerical scales and questionnaires. The aim of this work is to obtain relationships between these measurements. Most data recorded for LBP is numerical and range bound (intervalised or scaled), therefore presenting inherent fuzziness. To find relationships in the data, we have used a fuzzy association rule mining approach to identify correlations. Further, the use of fuzzy terms (linguistic terms) in the generated fuzzy rules helps to interpret the clinical outcome (readability). To show the applicability of this method, we have conducted experiments on a real LBP clinical dataset which indicate both valuable associations and understandable and interpretable rules. The results have been validated by an exercise and sports scientist.

Pair wise comparison (PC) is a well-established method to assist decision makers in estimating their preferences. This paper presents an open-source priority estimation tool (PriEsT) that has been developed to offer new features related to the PC method. PriEsT is able to assist decision makers (DMs) in revising their judgments based on different consistency measures and graphical aids. When inconsistency cannot be improved due to practical limitations, PriEsT offers a wide range of Pareto-optimal solutions based on multi-objective optimization, unlike other tools which offer only a single solution. DMs have the flexibility to select any of these non-dominated solutions according to their requirements. Using PriEsT, DMs can also perform Sensitivity Analysis to analyze how sensitive the solution is to changes in input data. The features of PriEsT have been demonstrated and evaluated via a real-world case study: an application to select the most appropriate Telecom infrastructure for rural areas. In the case study, the use of PriEsT has highlighted the presence of intransitive judgments in the acquired data. It has been found that correcting these judgments leads to a different ranking of the available alternatives. The generated solutions have also been tested for their sensitivity against small changes in the judgments.

Every company listed on the London Stock Exchange is classified into an industry sector based on its primary activity, however, it may be both more interesting and valuable to group similarly performing companies based on their historical stock price record over a long period of time. Using fuzzy clustering analysis with a correlation-based metric, we obtain a more insightful categorization of the companies into groups with fuzzy boundaries, giving arguably a more realistic and detailed view of their relationships. Once cluster analysis is performed, we analyze the behaviour of discovered groups in terms of the volatility of their returns using both standard deviation and exponentially weighted moving average. This approach has the potential to be of practical relevance in the context of diversified portfolio construction as it can detect fuzzy clusters of correlated stocks that have lower inter-cluster correlation, analyze their volatility and sample potentially less risky combination of assets.

We are constantly being told that we live in the Information Era - the Age of BIG data. It is clearly apparent that organizations need to employ data-driven decision making to gain competitive advantage. Processing, integrating and interacting with more data should make it better data, providing both more panoramic and more granular views to aid strategic decision making. This is made possible via Big Data exploiting affordable and usable Computational and Storage Resources. Many offerings are based on the Map-Reduce and Hadoop paradigms and most focus solely on the analytical side. Nonetheless, in many respects it remains unclear what Big Data actually is, current offerings appear as isolated silos that are difficult to integrate and/or make it difficult to better utilize existing data and systems. Paper addresses this lacunae by characterising the facets of Big Data and proposing a framework in which Big Data applications can be developed. The framework consists of three Stages and seven Layers to divide Big Data application into modular blocks. The aim is to enable organizations to better manage and architect a very large Big Data application to gain competitive advantage by allowing management to have a better handle on data processing.

This paper proposes a Bayesian association rule mining algorithm (BAR) which combines the Apriori association rule mining algorithm with Bayesian networks. Two interesting-ness measures of association rules: Bayesian confidence (BC) and Bayesian lift (BL) which measure conditional dependence and independence relationships between items are defined based on the joint probabilities represented by the Bayesian networks of association rules. BAR outputs best rules according to BC and BL. BAR is evaluated for its performance using two anonymized clinical phenotype datasets from the UCI Repository: Thyroid disease and Diabetes. The results show that BAR is capable of finding the best rules which have the highest BC, BL and very high support, confidence and lift.

OBJECTIVE: Identification of clinical events (eg, problems, tests, treatments) and associated temporal expressions (eg, dates and times) are key tasks in extracting and managing data from electronic health records. As part of the i2b2 2012 Natural Language Processing for Clinical Data challenge, we developed and evaluated a system to automatically extract temporal expressions and events from clinical narratives. The extracted temporal expressions were additionally normalized by assigning type, value, and modifier. MATERIALS AND METHODS: The system combines rule-based and machine learning approaches that rely on morphological, lexical, syntactic, semantic, and domain-specific features. Rule-based components were designed to handle the recognition and normalization of temporal expressions, while conditional random fields models were trained for event and temporal recognition. RESULTS: The system achieved micro F scores of 90% for the extraction of temporal expressions and 87% for clinical event extraction. The normalization component for temporal expressions achieved accuracies of 84.73% (expression's type), 70.44% (value), and 82.75% (modifier). DISCUSSION: Compared to the initial agreement between human annotators (87-89%), the system provided comparable performance for both event and temporal expression mining. While (lenient) identification of such mentions is achievable, finding the exact boundaries proved challenging. CONCLUSIONS: The system provides a state-of-the-art method that can be used to support automated identification of mentions of clinical events and temporal expressions in narratives either to support the manual review process or as a part of a large-scale processing of electronic health databases.

The key test for confidence in any association discovered within the medical domain is replication testing. That is, the ability of the association to be detected in independent populations. At the same time, in order to increase the likelihood of discovering statistically significant associations there is a clear need to increase the statistical power of any given study. A key methodology for increasing statistical power is through the use of as many subjects as possible that match a study's inclusion criteria. Thus many have attempted to merge data from multiple independent sources/sites/studies that contain the same inclusion criteria for subjects as a way of creating a much larger study with significantly more statistical power. For these approaches to work though data from multiple sites need to be made available to a single analysis. This practice is significantly limited by the need to respect legal and ethical requirements that are often complicated, ambiguous and inconsistent across different countries. The common approach to achieve merging of data is by sharing aggregated data rather than subject's personal data. Aggregated data however may still in some cases be reverse engineered, therefore traditionally cells within the aggregated data with small values were suppressed, and some or all of the aggregated data were perturbed in order to add noise inhibiting any attempts at identifying personal information of a specific person or sub-group in the original data. In this paper we study the effects of cell-suppression and perturbation on the results of the data analysis. Each approach is looked at by itself as well as in combination using the typical settings documented in the literature. The tests are based on a real dataset that looks for associations between phenotypes and genetic markers. This work is part of the Linked2Safety project that aims to dynamically interconnect distributed patients' data to better enable medical research efforts, whilst respecting patients' anonymity, as well as European and national legislation.

This paper presents a structure learning method for fuzzy systems following our previous work on a Structure Evolving Learning Method for Fuzzy Systems (SELM) and an Evolving Construction Scheme for Fuzzy Systems (ECSFS). Here we extend our previous work to a structure learning method for fuzzy systems which results in more concise systems. We analyse and compare the proposed concise structure learning strategies in terms of three aspects: (1) how to detect the splitting points for the structure learning process; (2) how to set a starting point for the fuzzy system; (3) how the proposed method is applied to Mamdani and TS fuzzy systems.

This paper presents a novel context-based feature for image encoding and object detection. The Oriented Gradient Context (OGC) descriptor represents the image in the context of different local area-based oriented gradient information. Both fine and coarse oriented gradients information about the image is captured, then different sizes of local areas with statistical oriented gradients are assembled into pair combinations to represent the gradient distribution context of the image. The features are comparatively simple but information-rich for utilization by classification algorithms. Based on the context information, the detection algorithm is relatively invariant to small shifts, translations of objects and changes in object appearance; even cases with partial occlusions and cluttered background are handled. The detection algorithm based on the proposed OGC features is shown to achieve good performance on pedestrian detection, comparable to other popular algorithms.

In this paper, we propose an Output-Constricted Clustering (OCC) algorithm for Radial Basis Function Neural Network (RBFNN) initialization. OCC first roughly partitions the output based on the required precision and then refinedly clusters data based on the input complexity within each output partition. The main contribution of the proposed clustering algorithm is that we introduce the concept of separability , which is a criterion to judge the suitability of the number of sub-clusters in each output partition. As a result, OCC is able to determine the proper number of sub-clusters with appropriate locations within each output partition by considering both input and output information. The resulting clusters from OCC are used to initialize RBFNN, with proper number and initial locations of for hidden neurons . As a result, RBFNN starting it's learning from a good point, is able to achieve better approximation performance than existing clustering methods for RBFNN initialization. This better performance is illustrated by a number of examples.

This paper presents an output-constrained clustering approach for fuzzy system identification and fuzzy granular system identification. The approach is unlike most existing clustering algorithms for structure identification of fuzzy systems, where the focus is on input or combined input-output clustering. The output-constrained clustering algorithm divides the output space into several partitions and each output partition is considered to be a constraint; then, input data are projected into clusters that are based on the input distribution constrained by the output partitions. By introducing the key concept of separability of a set of clusters within each output constraint, the proposed approach automatically finds an appropriate small and efficient number of clusters for each output constraint. To have an appropriate small and efficient number of clusters in each output constraint results in a more compact final system structure and better accuracy. This better performance is illustrated by experiments using benchmark function approximation and dynamic system identification.

OBJECTIVE: This study presents a system developed for the 2009 i2b2 Challenge in Natural Language Processing for Clinical Data, whose aim was to automatically extract certain information about medications used by a patient from his/her medical report. The aim was to extract the following information for each medication: name, dosage, mode/route, frequency, duration and reason. DESIGN: The system implements a rule-based methodology, which exploits typical morphological, lexical, syntactic and semantic features of the targeted information. These features were acquired from the training dataset and public resources such as the UMLS and relevant web pages. Information extracted by pattern matching was combined together using context-sensitive heuristic rules. MEASUREMENTS: The system was applied to a set of 547 previously unseen discharge summaries, and the extracted information was evaluated against a manually prepared gold standard consisting of 251 documents. The overall ranking of the participating teams was obtained using the micro-averaged F-measure as the primary evaluation metric. RESULTS: The implemented method achieved the micro-averaged F-measure of 81% (with 86% precision and 77% recall), which ranked this system third in the challenge. The significance tests revealed the system's performance to be not significantly different from that of the second ranked system. Relative to other systems, this system achieved the best F-measure for the extraction of duration (53%) and reason (46%). CONCLUSION: Based on the F-measure, the performance achieved (81%) was in line with the initial agreement between human annotators (82%), indicating that such a system may greatly facilitate the process of extracting relevant information from medical records by providing a solid basis for a manual review process.

This paper proposes an evolving-construction scheme for fuzzy systems (ECSFS). ECSFS begins with a simple fuzzy system and evolves its structure by adding more fuzzy terms and rules to achieve a better accuracy in a "greedy'' way. An interesting feature of ECSFS is that it is able to automatically locate mathematically meaningful points, such as the extremum and inflexion points of the approximated function one by one, and then adds fuzzy terms based on these points. Fuzzy systems with such extreme points, like their fuzzy terms, are more efficient than other fuzzy systems by using the same number of fuzzy rules. As a result, ECSFS often achieves a better accuracy for fuzzy-system identification compared with the previous methods when using the same number of fuzzy rules. A number of simulation results are given to illustrate the advantages of the proposed scheme.

Abstract Three parallel implementations of a divide‐and‐conquer search algorithm (called SUDA2) for finding minimal unique itemsets (MUIs) are compared in this paper. The identification of MUIs is used by national statistics agencies for statistical disclosure assessment. The first parallel implementation adapts SUDA2 to a symmetric multi‐processor cluster using the message passing interface (MPI), which we call an MPI cluster; the second optimizes the code for the Cray MTA2 (a shared‐memory, multi‐threaded architecture) and the third uses a heterogeneous ‘group’ of workstations connected by LAN. Each implementation considers the parallel structure of SUDA2, and how the subsearch computation times and sequence of subsearches affect load balancing. All three approaches scale with the number of processors, enabling SUDA2 to handle larger problems than before. For example, the MPI implementation is able to achieve nearly two orders of magnitude improvement with 132 processors. Performance results are given for a number of data sets. Copyright © 2009 John Wiley & Sons, Ltd.

OBJECTIVE The authors present a system developed for the Challenge in Natural Language Processing for Clinical Data-the i2b2 obesity challenge, whose aim was to automatically identify the status of obesity and 15 related co-morbidities in patients using their clinical discharge summaries. The challenge consisted of two tasks, textual and intuitive. The textual task was to identify explicit references to the diseases, whereas the intuitive task focused on the prediction of the disease status when the evidence was not explicitly asserted. DESIGN The authors assembled a set of resources to lexically and semantically profile the diseases and their associated symptoms, treatments, etc. These features were explored in a hybrid text mining approach, which combined dictionary look-up, rule-based, and machine-learning methods. MEASUREMENTS The methods were applied on a set of 507 previously unseen discharge summaries, and the predictions were evaluated against a manually prepared gold standard. The overall ranking of the participating teams was primarily based on the macro-averaged F-measure. RESULTS The implemented method achieved the macro-averaged F-measure of 81% for the textual task (which was the highest achieved in the challenge) and 63% for the intuitive task (ranked 7(th) out of 28 teams-the highest was 66%). The micro-averaged F-measure showed an average accuracy of 97% for textual and 96% for intuitive annotations. CONCLUSIONS The performance achieved was in line with the agreement between human annotators, indicating the potential of text mining for accurate and efficient prediction of disease statuses from clinical discharge summaries.

When applying gradient descent learning methods to hierarchical fuzzy systems, there is great difficulty in handling the intermediate variables introduced by the hierarchical structures, as the intermediate variables may go outside their definition domain that makes gradient descent learning invalid. To overcome this difficulty, this paper proposes a learning scheme that integrates a normalization process for intermediate variables into gradient descent learning. This ensures that gradient descent methods are applicable to, and correctly used for, learning general hierarchical fuzzy systems. Benchmark datasets are used to demonstrate the validity and advantages of the proposed learning scheme over other existing methods in terms of better accuracy, better transparency, and fewer fuzzy rules and parameters.

SUDA2 is a recursive search algorithm for minimal unique itemset detection. Such sets of items are formed via combinations of non-obvious attributes enabling individual record identification. The nature of SUDA2 allows work to be divided into non-overlapping tasks enabling parallel execution. Earlier work developed a parallel implementation for SUDA2 on an SMP cluster, and this was found to be several orders of magnitude faster than sequential SUDA2. However, if fixed-granularity parallel tasks are scheduled naively in the order of their generation, the system load tends to be imbalanced with little work at the beginning and end of the search. This paper investigates the effectiveness of variable-grained and dynamic work generation strategies for parallel SUDA2. These methods restrict the number of sub-tasks to be generated, based on the criterion of probable work size. The further we descend in the search recursion tree, the smaller the tasks become, thus we only select the largest tasks at each level of recursion as being suitable for scheduling. The revised algorithm runs approximately twice as fast as the existing parallel SUDA2 for finer levels of granularity when variable-grained work generation is applied. The dynamic method, performing level-wise task selection based on size, outperforms the other techniques investigated.

This paper proposes an incremental construction learning algorithm for identification of T-S Fuzzy Systems. The mechanism of the algorithm is that it is an error-reducing driven learning method. Beginning with a simplest T-S fuzzy system, the algorithm develops the system structure by adding more fuzzy terms and rules to reduce the model errors in a ‘greedy’ way. The main features of the proposed algorithm are that, firstly, it can automatically determines and controls the number and location of fuzzy terms needed by following the error-reducing driven evolving process to achieve the desired accuracy; secondly, it adds new fuzzy terms and rules by evenly distributing error to each sub-region aiming at an efficient set of fuzzy rules, thirdly, it uses triangular membership functions and the regular partitions in constructing T-S fuzzy systems and leads to identified T-S fuzzy system models with good transparency and interpretability and suitable for advanced stability analysis and design approaches such as piecewise Lyapounov methods. Two dynamical system identification examples are given to illustrate the advantages of the proposed algorithm.

BACKGROUND: Availability of information about transcription factors (TFs) is crucial for genome biology, as TFs play a central role in the regulation of gene expression. While manual literature curation is expensive and labour intensive, the development of semi-automated text mining support is hindered by unavailability of training data. There have been no studies on how existing data sources (e.g. TF-related data from the MeSH thesaurus and GO ontology) or potentially noisy example data (e.g. protein-protein interaction, PPI) could be used to provide training data for identification of TF-contexts in literature. RESULTS: In this paper we describe a text-classification system designed to automatically recognise contexts related to transcription factors in literature. A learning model is based on a set of biological features (e.g. protein and gene names, interaction words, other biological terms) that are deemed relevant for the task. We have exploited background knowledge from existing biological resources (MeSH and GO) to engineer such features. Weak and noisy training datasets have been collected from descriptions of TF-related concepts in MeSH and GO, PPI data and data representing non-protein-function descriptions. Three machine-learning methods are investigated, along with a vote-based merging of individual approaches and/or different training datasets. The system achieved highly encouraging results, with most classifiers achieving an F-measure above 90%. CONCLUSIONS: The experimental results have shown that the proposed model can be used for identification of TF-related contexts (i.e. sentences) with high accuracy, with a significantly reduced set of features when compared to traditional bag-of-words approach. The results of considering existing PPI data suggest that there is not as high similarity between TF and PPI contexts as we have expected. We have also shown that existing knowledge sources are useful both for feature engineering and for obtaining noisy positive training data.

This paper investigates the capabilities of hierarchical fuzzy systems to approximate functions on discrete input spaces. First, it is shown that any function on a discrete space has an arbitrary separable hierarchical structure and can be naturally approximated by hierarchical fuzzy systems. As a by-product of this result, a discrete version of Kolmogorov's theorem is obtained; second, it is proven that any function on a discrete space can be approximated to any degree of accuracy by hierarchical fuzzy systems with any desired separable hierarchical structure. That is, functions on discrete spaces can be approximated more simply and flexibly than those on continuous spaces; third, a hierarchical fuzzy system identification method is proposed in which human knowledge and numerical data are combined for system construction and identification. Finally, the proposed method is applied to the market condition performance modeling problem in site selection decision support and shows the better performance in both accuracy and interpretability than the regression and neural network approaches. In additions, the reason and mechanism why hierarchical fuzzy systems outperform regression and neural networks in this type of application are analyzed.

The ultimate purpose of a pedestrian-detection system (PDS) is to reduce pedestrian-vehicle-related injury. Most such systems tend to adopt expensive sensors, such as infrared devices, in expectation of better performance. In comparison, a low-cost optical-camera-based system has much potential practical value, including a greater detection range, and can easily be trained to detect other objects. However, such low-cost systems are difficult to design (e.g., little original information can be collected, and the scene is very complex). To address these problems, an effective and reliable classifier is needed. The classifier should have a proper structure, its features need to be well selected, and a large number of high-quality samples are necessary for training. In this paper, we present a low-cost PDS which only uses a single optical camera. We design a cascade classifier to achieve an effective and reliable detection. First, our system scans two sequential frames at each zoom scale with a sliding window. Second, with each window, both appearance and motion features are extracted. A well-trained cascade classifier, combining statistical learning with a decomposed support-vector-machine classifier, then determines whether the window contains a human body. At the same time, to provide as much information as possible about the pedestrian, a small-scale weighted template tree trained by a coevolutionary algorithm is adopted to identify each pedestrian's direction, and the distance of each from the vehicle is also provided using an estimation algorithm. During the training procedure, we select key features by using the AdaBoost algorithm and a large number of high-quality samples. Experimental results demonstrate that the system is suitable for pedestrian detection in city traffic: The detection speed is more than 10 ft/s, the detection rate reaches 80%, and the false positive rate is no more than 0.30/00.

Short-term load forecasting has become increasingly important since the rise of the competitive energy markets and has become one of the major areas of research in recent years. Toward to this important topic, this paper proposes a new approach- the self-organizing fuzzy neural network (SOFNN) modeling method for the short-term load forecasting. The main advantages of this approach are that, firstly, it is very user friendly as SOFNN can automatically determine the model structure and identify the model parameters without requiring the in-depth knowledge about fuzzy systems and neural networks; secondly, it provides the excellent forecasting accuracy. Applying this approach based on the real data, the achieved average mean absolute percentage error (MAPE) for load forecasting is less than 1%.

Rough set feature selection (RSFS) can be used to improve classifier performance. RSFS removes redundant attributes whilst keeping important ones that preserve the classification power of the original dataset. The feature subsets selected by RSFS are termed reducts. The intersection of all reducts is termed the core. As RSFS works on discrete attributes only, for real-valued datasets discretization of the real attributes is performed before RSFS. The core size of the discretized datasets is determined by the discretization process. Previous work has shown that the core size of the discretized dataset critically affects the performance of RSFS. This paper proposes a type of discretization termed core-generating approximate minimum entropy discretization (C-GAME) which selects a set of minimum entropy cuts capable of generating discrete data with nonempty cores. The paper defines C-GAME and then models it as a constraint satisfaction optimization problem which is solved using the branch and bound algorithm. Experiments have been performed on 2 datasets from the UCI database to investigate the performance of C-GAME as a pre-processing step for RSFS. Results show that, for these datasets, C-GAME outperforms both the recursive minimal entropy partition discretization method (RMEP) and the original decision trees without feature selection.

Clustering algorithms are often used for fuzzy system identification. However, most clustering algorithms do not consider the outputs for clustering. In addition, these algorithms do not consider how to obtain the optimal number of clusters. Without the optimal number of clusters, the final set of clusters may be inappropriate. To address this, this paper presents an Input-Output Clustering (IOC) algorithm to determine both the correct number of clusters and the appropriate location for them by considering both inputs and outputs. The proposed algorithm, when used for fuzzy system identification, achieves better performance than existing clustering methods. This performance is illustrated by examples of function approximation and dynamic system identification.

Standard fuzzy systems suffer the "curse of dimensionality" which has become the bottleneck when applying fuzzy systems to solve complex and high dimensional application problems. This curse of dimensionality results in a larger number of fuzzy rules which reduces the transparency of fuzzy systems. Furthermore too many rules also reduce the generalization capability of fuzzy systems. Hierarchical fuzzy systems have emerged as an effective alternative to overcome this curse of dimensionality and have attracted much attention. However, research on learning methods for hierarchical fuzzy systems and applications is rare. In this paper, we propose a scheme to construct general hierarchical fuzzy systems based on the gradient-descent method. To show the advantages of the proposed method (in terms of accuracy, transparency, generalization capability and fewer rules), this method is applied to a function approximation problem and the result is compared with those obtained by standard (flat) fuzzy systems.

This paper investigates the stabilization of conventional nonlinear systems by fuzzy control approach. Firstly, it is shown that the class of nonlinear systems whose stabilization problem can be solved by the fuzzy control approach available today based on T-S fuzzy control models is affine nonlinear systems as this is the only class of nonlinear systems which can be approximated to any degree of accuracy by T-S fuzzy control models; secondly, it shows that the stabilization problem of an affine nonlinear system can be solved as a robust stabilization problem of its T-S fuzzy approximator with approximation error bound as the system uncertainty bound; thirdly, for an affine nonlinear system, an approximation scheme is proposed to construct its T-S approximator and the corresponding approximation error bound is obtained; finally, it discusses briefly how to solve the stabilization problem of an affine nonlinear system by solving its corresponding robust stabilization problem based on its T-S fuzzy approximator and approximation error bound by Lyapunov's method.

Gilbert algorithm is a very popular algorithm in collision detection in robotics and also in classification in pattern recognition. However, the major drawback of Gilbert algorithm is that in many cases it becomes very slow as it approaches the final solution and the vertices selection vibrates in these cases. In this paper: a) It is proven theoretically that when the selection of vertices vibrates among several points, the algorithm will converge to the hyperplane determined by these points. b) Based on the above results, an improved Gilbert algorithm for computing the distance between two convex polytopes is presented. The algorithm can avoid the slow convergence of the original one. Numerical simulation results demonstrate the effectiveness and advantage of the improved algorithm

We created a proximity measure, called Proximity Within Paragraph (PWP), which is based on the concept of value assignment to queried words, grouped by associated ideas within paragraphs. Based on the WT10G dataset, a test system comprising three test sets and fifty queries were constructed to evaluate the effectiveness of PWP by comparing it with the existing method: Minimum Distance Between Queried Pairs. A further experiment combines the scores obtained from both methods and the results suggest that the combination can significantly improve the effectiveness.

This paper investigates the approximation capabilities of hierarchical hybrid systems, which are motivated by research in hierarchical fuzzy systems, hybrid intelligent systems, and modeling of model partly known systems. For a function (system) with known hierarchical structure (i.e., one that can be represented as a composition of some simpler and lower dimensional subsystems), it is shown that hierarchical hybrid systems have the structure approximation capability in the sense that such a hybrid approximation scheme can approximate both the overall system and all the subsystems to any desired degree of accuracy. For a function (system) with unknown hierarchical structure, Kolmogorov's theorem is used to construct the hierarchical structure of the given function (system). It is then shown that hierarchical hybrid systems are universal approximators

This paper discusses the capabilities of standard hierarchical fuzzy systems to approximate continuous functions with natural hierarchical structure. The separable approximation property of hierarchical fuzzy systems is proved, that is, the construction of a hierarchical fuzzy system with required approximation accuracy can be achieved by the separate construction of each sub-system with required approximation accuracy. This property provides a simple method to construct hierarchical fuzzy systems for function approximation. Based on the separable approximation property, it is further proved the structure approximation property of hierarchical fuzzy systems

Derived from practical application in location analysis and pricing, and based on the approach of hierarchical structure analysis of continuous functions, this paper investigates the approximation capabilities of hierarchical fuzzy systems. By first introducing the concept of the natural hierarchical structure, it is proved that continuous functions with natural hierarchical structure can be naturally and effectively approximated by hierarchical fuzzy systems to overcome the curse of dimensionality in both the number of rules and parameters. Then, based on Kolmogorov's theorem, it is shown that any continuous function can be represented as a superposition of functions with the natural hierarchical structure and can then be approximated by hierarchical fuzzy systems to achieve the universal approximation property. Further, the conditions under which the hierarchical fuzzy approximation is superior to the standard fuzzy approximation in overcoming the curse of dimensionality are analyzed

Based on the approach of hierarchical structure analysis of continuous functions, this paper discusses the approximation capacities of hierarchical Takagi-Sugeno fuzzy systems. By first introducing the concept of the natural hierarchical structure, it is proved that continuous functions with the natural hierarchical structure can be naturally and effectively approximated by hierarchical fuzzy systems to overcome the curse of dimensionality in both the number of rules and the number of parameters. Then, based on the Kolmogorov's theorem, it is shown that any continuous function can be represented as a superposition of functions with natural hierarchical structure and can then be approximated by hierarchical Takagi-Sugeno fuzzy systems to achieve the universal approximation property.

This paper reports on our experiences with using the emerging Web service technologies and tools to create a demonstration information broker system as part of our research into information management in a distributed environment. To provide a realistic context, we chose to study the use of information in the healthcare domain, and this context sets some challenging parameters and constraints for our research and for the demonstration system. In this paper, we both report on the extent to which existing Web service technologies have proved to be mature enough to meet these requirements, and also assess their current limitations.

We address the problem of large-scale data integration, where the data sources are unknown at design time, are from autonomous organisations, and may evolve. Experiments are described involving a demonstrator system in the field of health services data integration within the UK. Current Web services technology has been used extensively and largely successfully in these distributed prototype systems. The work shows that Web services provide a good infrastructure layer, but integration demands a higher level "broker" architectural layer; the paper identifies eight specific requirements for such an architecture that have emerged from the experiments, derived from an analysis of shortcomings which are collectively due to the static nature of the initial prototype. The way in which these are being met in the current version in order to achieve a more dynamic integration is described.

Support vector machines (SVM) decomposition methods were proposed to solve high dimensional and/or large data classification problems. Two major decomposition algorithms: Karush-kuhn-Tucker (KKT) condition based algorithm, and 'Joachims' decomposition algorithm are popularly adopted. In this paper, both these two decomposition methods are analyzed and applied into face recognition with three basic mapping kernels. Numerical results showed that: a) face recognition with SVM performs better accuracy than other existed methods; b) the decomposition methods can perform face recognition efficiently; c) Joachims' decomposition method has better accuracy than that of decomposition algorithm based on KKT condition; d) linear kernel can provide much higher recognition accuracy than polynomial and slightly better accuracy than Gaussian radial based function (RBF) kernel; Also due to the fact that the linear kernel method is much simpler than others, it is most suitable for face recognition.

Effective use of heterogeneous, distributed information in a coherent, integrated fashion has long been a "holy grail". The UK Health and Social care domain is an example where a global view is needed to facilitate decision making, whilst having ethical and legal concerns. The IBHIS project aims to provide an integrated broker that enables coherent use of a set of distributed, heterogeneous data sources, whilst ensuring trustworthiness and audit. This paper presents a service-oriented federated architecture for the IBHIS broker.

Location analysis decisions are interrelated and should be made within a single decision-making framework. A framework within which a number of location strategies can be placed is presented. Location-allocation models are improved in two ways: 1) the allocation rule is developed to more accurately reflect customer choice processes; and 2) the objective function is developed to incorporate future changes. Computational support for this framework is described.

Attribute-Oriented Induction (AOI) is a data mining technique that produces simplified descriptive patterns. Classical AOIuses a predictive strategy to determine distinct values of an attribute but generalises attributes indiscriminately i.e. the value ‘ANY’ is replaced like any other value without restrictions. AOI only produces interesting rules by using interior concepts of attribute hierarchies. The COMPARE algorithm that integrates predictive and lookahead methods and of order complexity O (np), where n and p are input and generalised tuples respectively, is introduced. The latter method determines distinct values of attribute clusters and greatest number of attribute values with a ‘common parent’ (except parent ‘ANY’). When generating rules, a rough set approach to eliminate redundant attributes is used leading to more interesting multiple-level rules with fewer ‘ANY’ values than classical AOI.

This paper presents the design and development of a co-ordination framework that enables a distributed, heterogeneous environment to be used as a single integrated system. The design has been motivated by the requirements of financial applications: in particular by a computationally intensive actuarial system that for large schemes requires computing and data resources. The framework is structured into two levels: an interprocess communication layer for application programmers, and an underlying communication management layer. The envisaged distributed environment includes MPPs, SMPs and workstations, connected via LANs and WANs. Copyright © 1999 John Wiley & Sons, Ltd.

TSQL2 is a query language designed for temporal databases. In TSQL2, the GROUP BY clause has a temporal grouping property. In temporal grouping, the time line of each attribute value is partitioned into several sections, and aggregate functions are computed for each time partition. This paper describes two parallel algorithms, data partition and group partition, which compute temporal aggregates over a network of workstations. In the group partition scheme, each workstation maintains the entire aggregate tree for some attribute values selected by the GROUP BY clause. Thus, some workstations may be overloaded while others are idle for most of the time. In the data partition scheme, all the workstations participate in constructing the aggregate trees in the first phase of the scheme. Thus the load is evenly distributed across the workstations in the system in the first phase of the scheme. However, before the second phase starts, workstations must exchange the aggregate trees generated at the first phase. A simulator has been used to test the performance of the two algorithms. The results show that the performance of algorithm group partition is slightly better than data partition.

Association rules are used to express "interesting" relationships between items of data in a standard enterprise database. In a temporal database, each tuple is given a start and an end time indicating the period during which the information recorded in the tuple is valid. With a temporal database, we may wish to discover relationships between items which satisfy certain timing constraints. Existing algorithms for mining association rules cannot be applied to temporal databases directly. This is because, in the existing algorithms, if an itemset is supported by a tuple, the tuple must contain all the items in the itemset. For temporal databases, an itemset, e.g. {A,B}, is supported as long as all the items in {A,B} are contained in a set of tuples which satisfy certain timing constraint (e.g. the duration of the tuples containing A and B overlap each other). In this paper, an algorithm for mining association rules in temporal databases is described. The algorithm allows (a) the itemsets to contain composite items, and (b) the timing constraint on the tuples to be specified by the users.

This paper reports on an ongoing investigation into the applicability of parallel systems in the processing of financial information. For the past five years a wide spectrum of issues has been investigated; this paper attempts to draw together this work as a coherent whole, and discusses past, present and planned activity. A number of case studies applying parallel systems to typical financial applications are described and, where appropriate, results presented. The commonality between the case studies is also discussed.

Abstract The Flagship Project 1 was a research collaboration between the University of Manchester, Imperial College London and International Computers Ltd. The project was unusual in that it aimed to produce a complete computing system based on a declarative programming style. Three areas of a declarative system were addressed: (1) programming languages and programming environments; (2) the machine architecture and computational models; and (3) the software environment. This overview paper discusses each of these areas, the intention being to present the project as a coherent whole.

Abstract The Flagship system is a graph reduction machine having a distributed physical architecture. Although Flagship sits firmly in the declarative world, explicit state is supported to express the behaviour of the operating system. This state not only has to be isolated from the declarative aspects of the Flagship machine, but also has to be supported with respect to distribution. The mechanisms provided for maintaining consistency of state are discussed with respect to atomic actions at levels in the Flagship system. This approach is used to demonstrate how the software environment was supported by the basic execution mechanism of the machine. The nature, construction and creation of system software components are described and the structure of the system software is discussed with particular reference to optimising access to distributed resources.

The authors report on the use of formal methods for the development of parallel operating systems for two experimental declarative systems over a five-year period. A common specification approach has evolved as part of the development of these two very different systems: one being for a parallel graph reduction machine and written in a functional language enhanced with state-based objects, the other was written in C/sup ++/. A brief overview of each system is given before concentrating on the use of formal methods. A description is given of how both a technique for formally specifying sequential systems (VDM) and a technique for specifying concurrent systems (temporal logic) have been used together. In both cases, the issue of verification is addressed.>