Mining for association rules in market basket data has proved a fruitful area of research. Measures such as conditional probability (confidence) and correlation have been used to infer rules of the form "the existence of item A implies the existence of item B." However, such rules indicate only a statistical relationship between A and B. They do not specify the nature of the relationship: whether the presence of A causes the presence of B, or the converse, or some other attribute or phenomenon causes both to appear together. In applications, knowing such causal relationships is extremely useful for enhancing understanding and effecting change. While distinguishing causality from correlation is a truly difficult problem, recent work in statistics and Bayesian learning provide some avenues of attack. In these fields, the goal has generally been to learn complete causal models, which are essentially impossible to learn in large-scale data mining applications with a large number of variab...

Abstract. Semantically heterogeneous and distributed data sources are quite common in several application domains such as bioinformatics and security informatics. In such a setting, each data source has an associated ontology. Different users or applications need to be able to query such data sources for statistics of interest (e.g., statistics needed to learn a predictive model from data). Because no single ontology meets the needs of all applications or users in every context, or for that matter, even a single user in different contexts, there is a need for principled approaches to acquiring statistics from semantically heterogeneous data. In this paper, we introduce ontology-extended data sources and define a user perspective consisting of an ontology and a set of interoperation constraints between data source ontologies and the user ontology. We show how these constraints can be used to derive mappings from source ontologies to the user ontology. We observe that most of the learning algorithms use only certain statistics computed from data in the process of generating the hypothesis that they output. We show how the ontology mappings can be used to answer statistical queries needed by algorithms for learning classifiers from data viewed from a certain user perspective. The resulting algorithms offer a powerful approach to data-driven knowledge acquisition over the Semantic Web.

Recent advances in computing, communications, and digital storage technologies, together with development of high throughput data acquisition technologies have made it possible to gather and store large volumes of data in digital form. These developments have resulted in unprecedented opportunities for large-scale data-driven knowledge acquisition with the potential for fundamental gains in scientific understanding (e.g., characterization of macromolecular structure-function relationships in biology) in many data-rich domains. In such applications, the data sources of interest are typically physically distributed, semantically heterogeneous and autonomously owned and operated, which makes it impossible to use traditional machine learning algorithms for knowledge acquisition. However, we observe that most of the learning algorithms use only certain statistics computed from data in the process of generating the hypothesis that they output and we use this observation to design a general strategy for transforming traditional algorithms for learning from data into algorithms for learning from distributed data. The resulting algorithms are provably exact in that the classifiers produced by them are identical to those obtained by the corresponding algorithms in the centralized setting (i.e., when all of the data is available in a central location) and they compare favorably to their centralized counterparts in terms of time and communication complexity. To deal with the semantical heterogeneity problem, we introduce ontology-extended data sources and define a user perspective consisting of an ontology and a set of interoperation constraints between data source ontologies and the user ontology. We show how these constraints can be used to define mappings and conversion functions needed to answer statistical queries from semantically heterogeneous data viewed from a certain user perspective. That is further used to extend our approach for learning from distributed data into a theoretically sound approach to learning from semantically heterogeneous data. The work described above contributed to the design and implementation of AirlDM, a collection of data source independent machine learning algorithms through the means of sufficient statistics and data source wrappers, and to the design of INDUS, a federated, query-centric system for knowledge acquisition from distributed, semantically heterogeneous, autonomous data sources.

implications for drug resistance

Concepts like marginal probability, changes in probability, probability propagation and cause-effect relationships are important when reasoning about causality and uncertainty.

Abstract not found

Abstract. Bayesian Belief Networks (BBNs) have been suggested as a suitable representation and inference mechanism for learner models (Reye, 1998). Having the goal to construct inspectable Bayesian learner models, we have encountered two problems. The first is the problem of visualization of BBNs and the second is the problem of using BBNs to model changes in knowledge state over time due to gradual learning and forgetting. Both of these problems are being addressed in current research at the ARIES Laboratory and both are briefly discussed in this paper. 1

this paper we present a method for answering relationship questions, as posed for example in the spring of 2003 evaluation exercise of the AQUAINT program, which has funded this research

Bayesian networks are a popular mechanism for dealing with uncertainty in complex situations. They are a fundamental probabilistic representation mechanism that subsumes a great variety of other stochastic modeling methods, such as hidden Markov models, stochastic dynamic systems. Bayesian networks, in principle, make it possible to build large, complex stochastic models from standard components. Development methodologies for Bayesian networks have been introduced based on software engineering methodologies. However, this is complicated by the significant differences between the crisp, logical foundations of modern software and the fuzzy, empirical nature of stochastic modeling. Conversely, software engineering would benefit from better integration with Bayesian networks, so that uncertainty and stochastic inference can be introduced in a more systematic and formal manner than it is now. In this paper, Bayesian networks and stochastic inference are briefly introduced, and the development of Bayesian networks is compared with the development of objectoriented software. The challenges involved in Bayesian network development are then discussed.

Decision analysis is an important part of development processes, and decision analysis techniques have been very highly developed. However, the connection between a particular decision analysis and the process within which it occurs is informal at best. Consequently, it can be difficult to revisit or reuse a decision analysis at a later time or for another purpose. Once a decision has been made, the analysis that was performed is often discarded, and even if it is retained, the context within which the analysis was performed is not documented sufficiently for the analysis to be easily revisited when the situation changes or to be reused for similar decisions. In this position paper we propose formal annotation of decision analyses using Semantic Web technologies, and the development of a formal process model using the Eclipse Process Framework (EPF) to document the roles, artifacts and processes that are used by decision analyses. Formal annotation also makes it possible to reason about the analysis and process, allowing one to check consistency and to answer queries. The use of EPF makes it possible to integrate the decision analysis process as a component in the process where it is used, and to reuse the decision analysis in other contexts.