Rattikorn Hewett

Tumor classification ranking from microarray data.

BackgroundGene expression profiles based on microarray data are recognized as potential diagnostic indices of cancer. Molecular tumor classifications resulted from these data and learning algorithms have advanced our understanding of genetic changes associated with cancer etiology and development. However, classifications are not always perfect and in such cases the classification rankings (likelihoods of correct class predictions) can be useful for directing further research (e.g., by deriving inferences about predictive indicators or prioritizing future experiments). Classification ranking is a challenging problem, particularly for microarray data, where there is a huge number of possible regulated genes with no known rating function. This study investigates the possibility of making tumor classification more informative by using a method for classification ranking that requires no additional ranking analysis and maintains relatively good classification accuracy.ResultsMicroarray data of 11 different types and subtypes of cancer were analyzed using MDR (Multi-Dimensional Ranker), a recently developed boosting-based ranking algorithm. The number of predictor genes in all of the resulting classification models was at most nine, a huge reduction from the more than 12 thousands genes in the majority of the expression samples. Compared to several other learning algorithms, MDR gives the greatest AUC (area under the ROC curve) for the classifications of prostate cancer, acute lymphoblastic leukemia (ALL) and four ALL subtypes: BCR-ABL, E2A-PBX1, MALL and TALL. SVM (Support Vector Machine) gives the highest AUC for the classifications of lung, lymphoma, and breast cancers, and two ALL subtypes: Hyperdiploid > 50 and TEL-AML1. MDR gives highly competitive results, producing the highest average AUC, 91.01%, and an average overall accuracy of 90.01% for cancer expression analysis.ConclusionUsing the classification rankings from MDR is a simple technique for obtaining effective and informative tumor classifications from cancer gene expression data. Further interpretation of the results obtained from MDR is required. MDR can also be used directly as a simple feature selection mechanism to identify genes relevant to tumor classification. MDR may be applicable to many other classification problems for microarray data.

Distributing Intelligence within an Individual

Distributed artificial intelligence (DAI) refers to systems in which decentralized, cooperative agents work synergistically to perform a task. Alternative DAI models resemble particular biological or social systems, such as teams, contract nets, or societies. Our DAI model resembles a single individual, characterized by adaptability, versatility, and coherence. The proposed DAI architecture comprises a hierarchy of loosely coupled agents for specific perception, action, and reasoning functions, all operating under the supervision of a top-level control agent. We demonstrate the proposed architecture in the Guardian system for intensive-care monitoring.

On modeling software defect repair time

The ability to predict the time required to repair software defects is important for both software quality management and maintenance. Estimated repair times can be used to improve the reliability and time-to-market of software under development. This paper presents an empirical approach to predicting defect repair times by constructing models that use well-established machine learning algorithms and defect data from past software defect reports. We describe, as a case study, the analysis of defect reports collected during the development of a large medical software system. Our predictive models give accuracies as high as 93.44%, despite the limitations of the available data. We present the proposed methodology along with detailed experimental results, which include comparisons with other analytical modeling approaches.

Host-Centric Model Checking for Network Vulnerability Analysis

Research has successfully applied model checking, a formal verification technique, to automatically generate chains of vulnerability exploits that an attacker can use to reach his goal. Due to the combinatorial explosion of the chain generation problem space, model checkers do not scale well to networks containing a large number of hosts. This paper proposes a methodology that uses a host-centric modeling approach together with a monotonicity assumption to alleviate the scalability problem of model checkers. We describe the proposed approach, its limitations, and show how it can reduce the time complexity of chain generation to a quadratic polynomial of the number of hosts, both theoretically and empirically. We also compare its advantages over similar customized graph-based approaches.

Scalable Optimized Composition of Web Services with Complexity Analysis

This paper addresses a fundamental issue of web service composition. We present a simple but powerful conceptual model that leads to a scalable approach to automatically constructing a composite web service to meet its requirements by using as few services as possible. Our approach is based on a state space model that has a monotone property to allow efficient search along with efficient algorithms for pruning and simple parallelization. We provide both empirical and theoretical analyses of the proposed approach and show that it has time complexity of O(n^2), for a repository with n services. However, the approach takes linear time for sequential compositions when service applicability is performed by service discovery and thus, it is shown to give asymptotically optimal performance. Although optimality in the number of services deployed is not guaranteed, our experiments on public benchmark data sets show correct optimized solutions 100% of the time, with a reduction in the average running time, compared to a well-performed planning-based system, of better than 35% over 207 composition problems.

Automated Test Order Generation for Software Component Integration Testing

The order in which software components are tested can have a significant impact on the number of stubs required during component integration testing. This paper presents an efficient approach that applies heuristics based on a given software component test dependency graph to automatically generate a test order that requires a (near) minimal number of test stubs. Thus, the approach reduces testing effort and cost. The paper describes the proposed approach, analyses its complexity and illustrates its use. Comparison with three well known graph-based approaches, for a real-world software application, shows that only the classic Le Traon et al.’s approach and ours give an optimal number of stubs. However, experiments on randomly simulated dependency models with 100 to 10,000 components show that our approach has a significant performance advantage with a reduction in the average running time of 96.01%.

Restructuring decision tables for elucidation of knowledge

Decision tables are widely used in many knowledge-based and decision support systems. They allow relatively complex logical relationships to be represented in an easily understood form and processed efficiently. This paper describes second-order decision tables (decision tables that contain rows whose components have sets of atomic values) and their role in knowledge engineering to: (1) support efficient management and enhance comprehensibility of tabular knowledge acquired by knowledge engineers, and (2) automatically generate knowledge from a tabular set of examples. We show how second-order decision tables can be used to restructure acquired tabular knowledge into a condensed but logically equivalent second-order table. We then present the results of experiments with such restructuring. Next, we describe SORCER, a learning system that induces second-order decision tables from a given database. We compare SORCER with IDTM, a system that induces standard decision tables, and a state-of-the-art decision tree learner, C4.5. Results show that in spite of its simple induction methods, on the average over the data sets studied, SORCER has the lowest error rate.

Automated Negotiations in Web Service Procurement

Any Web service procurement involves complex negotiations on non-functional requirements in order to obtain an integrative solution for both the provider and consumer perspectives. Software that facilitates automated Web service procurement negotiations is valuable not only for the consumers and providers to continuously bargain for their customizations and tailor their offerings, but also to discover overlooked alternative solutions and to maintain documented rationales for future references and reuse. To date, software frameworks and mechanisms that provide flexible environments to support automated negotiations in web service procurement are lagging. This paper presents a tradeoff-based automated negotiation approach to support Web service procurement. The approach employs an iterative tradeoff mechanism for evaluating opponents offers and generating counter-offers of mutual gain based on selected quality of service parameters. The paper presents the negotiation approach in details along with an illustrated case study of negotiation in Web service procurement for a medical insurance company.

A language and architecture for efficient blackboard systems

The authors present a low-level language for blackboard systems. They also present efficient blackboard activation and agenda maintenance mechanisms. The reason why RETE-like pattern-matching networks are not appropriate for blackboard systems is explained. Instead, activation demons provide an efficient mechanism for agenda maintenance. The results show a significant speedup in agenda maintenance and execution when using the new mechanism in the BB1 blackboard architecture.<<ETX>>

Security Analysis of Role-based Separation of Duty with Workflows

Role-based access control (RBAC) is the most predominant access control model in todays security management due to its ability to simplify authorization, and flexibility to specify and enforce protection policies. In RBAC, Separation of Duty (SoD) constrains user role authorization to protect sensitive information from frauds due to conflicts of interests. SoD constraints are commonly defined by mutually exclusive roles (MER) (e.g., bank teller and auditor). This paper proposes practical computational techniques for analyzing SoD by integrating workflows of the enterprise processes into the RBAC framework. Specifically, we present 1) an algorithm for generating MER to enforce SoD, and 2) a verification algorithm to check if a given RBAC state (role authorization and user-role assignments) satisfies a given type of SoD constraint or not. The paper discusses the details of the approach and illustrates its use in a loan application domain.