Instantiation Levels Research Articles

Learning the Finer Things: Bayesian Structure Learning at the Instantiation Level

Successful machine learning methods require a trade-off between memorization and generalization. Too much memorization and the model cannot generalize to unobserved examples. Too much over-generalization and we risk under-fitting the data. While we commonly measure their performance through cross validation and accuracy metrics, how should these algorithms cope in domains that are extremely under-determined where accuracy is always unsatisfactory? We present a novel probabilistic graphical model structure learning approach that can learn, generalize and explain in these elusive domains by operating at the random variable instantiation level. Using Minimum Description Length (MDL) analysis, we propose a new decomposition of the learning problem over all training exemplars, fusing together minimal entropy inferences to construct a final knowledge base. By leveraging Bayesian Knowledge Bases (BKBs), a framework that operates at the instantiation level and inherently subsumes Bayesian Networks (BNs), we develop both a theoretical MDL score and associated structure learning algorithm that demonstrates significant improvements over learned BNs on 40 benchmark datasets. Further, our algorithm incorporates recent off-the-shelf DAG learning techniques enabling tractable results even on large problems. We then demonstrate the utility of our approach in a significantly under-determined domain by learning gene regulatory networks on breast cancer gene mutational data available from The Cancer Genome Atlas (TCGA).

Automated Network Service Scaling in NFV: Concepts, Mechanisms and Scaling Workflow

Next-generation systems are anticipated to be digital platforms supporting innovative services with rapidly changing traffic patterns. To cope with this dynamicity in a cost-efficient manner, operators need advanced service management capabilities such as those provided by NFV. NFV enables operators to scale network services with higher granularity and agility than today. For this end, automation is key. In search of this automation, the European Telecommunications Standards Institute (ETSI) has defined a reference NFV framework that make use of model-driven templates called Network Service Descriptors (NSDs) to operate network services through their lifecycle. For the scaling operation, an NSD defines a discrete set of instantiation levels among which a network service instance can be resized throughout its lifecycle. Thus, the design of these levels is key for ensuring an effective scaling. In this article, we provide an overview of the automation of the network service scaling operation in NFV, addressing the options and boundaries introduced by ETSI normative specifications. We start by providing a description of the NSD structure, focusing on how instantiation levels are constructed. For illustrative purposes, we propose an NSD for a representative NS. This NSD includes different instantiation levels that enable different ways to automatically scale this NS. Then, we show the different scaling procedures the NFV framework has available, and how it may automate their triggering. Finally, we propose an ETSI-compliant workflow to describe in detail a representative scaling procedure. This workflow clarifies the interactions and information exchanges between the functional blocks in the NFV framework when performing the scaling operation.

A conceptual framework for large-scale ecosystem interoperability and industrial product lifecycles

One of the most significant challenges in information system design is the constant and increasing need to establish interoperability between heterogeneous software systems at increasing scale. The automated translation of data between the data models and languages used by information ecosystems built around official or de facto standards is best addressed using model-driven engineering techniques, but requires handling both data and multiple levels of metadata within a single model. Standard modelling approaches are generally not built for this, compromising modelling outcomes. We establish the SLICER conceptual framework built on multilevel modelling principles and the differentiation of basic semantic relations (such as specialisation, instantiation, specification and categorisation) that dynamically structure the model. Moreover, it provides a natural propagation of constraints over multiple levels of instantiation. The presented framework is novel in its flexibility towards identifying the multilevel structure, the differentiation of relations often combined in other frameworks, and a natural propagation of constraints over multiple levels of instantiation.

Run-time architectural modeling for future internet applications

Interoperability, flexibility, and adaptability are key requirements of future internet applications. Convergence of contents, services, things, and networks can be the cornerstone to fulfill these requirements. Such rich and composite sources of data and processing capabilities call for a structured and formal approach that manages and capitalizes heterogeneous information. This paper proposes an approach to the run-time composition of software system architectures, aimed at addressing goals revealed at runtime. The approach is grounded on a graph model characterized by two control levels: a metamodeling and an instantiation level. At metamodeling level, the graph describes facts that may occur in a scenario of interest, processes triggered by facts, and technologies available to execute processes. The actual occurrence of facts, together with the deriving processes and technologies, is managed at instantiation level, with reference to an application-specific model. In particular, the paper proposes an algorithm that determines an optimal way to manage a change in the run-time environment, by finding a minimum cost path in the model. The usefulness of the proposed approach and its applicability to actual scenarios have been validated in an example smart home environment.

Dual deep modeling: multi-level modeling with dual potencies and its formalization in F-Logic

An enterprise database contains a global, integrated, and consistent representation of a company’s data. Multi-level modeling facilitates the definition and maintenance of such an integrated conceptual data model in a dynamic environment of changing data requirements of diverse applications. Multi-level models transcend the traditional separation of class and object with clabjects as the central modeling primitive, which allows for a more flexible and natural representation of many real-world use cases. In deep instantiation, the number of instantiation levels of a clabject or property is indicated by a single potency. Dual deep modeling (DDM) differentiates between source potency and target potency of a property or association and supports the flexible instantiation and refinement of the property by statements connecting clabjects at different modeling levels. DDM comes with multiple generalization of clabjects, subsetting/specialization of properties, and multi-level cardinality constraints. Examples are presented using a UML-style notation for DDM together with UML class and object diagrams for the representation of two-level user views derived from the multi-level model. Syntax and semantics of DDM are formalized and implemented in F-Logic, supporting the modeler with integrity checks and rich query facilities.

Subjectivity detection in spoken and written conversations

AbstractIn this work we investigate four subjectivity and polarity tasks on spoken and written conversations. We implement and compare several pattern-based subjectivity detection approaches, including a novel technique wherein subjective patterns are learned from both labeled and unlabeled data, using n-gram word sequences with varying levels of lexical instantiation. We compare the use of these learned patterns with an alternative approach of using a very large set of raw pattern features. We also investigate how these pattern-based approaches can be supplemented and improved with features relating to conversation structure. Experimenting with meeting speech and email threads, we find that our novel systems incorporating varying instantiation patterns and conversation features outperform state-of-the-art systems despite having no recourse to domain-specific features such as prosodic cues and email headers. In some cases, such as when working with noisy speech recognizer output, a small set of well-motivated conversation features performs as well as a very large set of raw patterns.

Solving quantified verification conditions using satisfiability modulo theories

First-order logic provides a convenient formalism for describing a wide variety of verification conditions. Two main approaches to checking such conditions are pure first-order automated theorem proving (ATP) and automated theorem proving based on satisfiability modulo theories (SMT). Traditional ATP systems are designed to handle quantifiers easily, but often have difficulty reasoning with respect to theories. SMT systems, on the other hand, have built-in support for many useful theories, but have a much more difficult time with quantifiers. One clue on how to get the best of both worlds can be found in the legacy system Simplify which combines built-in theory reasoning with quantifier instantiation heuristics. Inspired by Simplify and motivated by a desire to provide a competitive alternative to ATP systems, this paper describes a methodology for reasoning about quantifiers in SMT systems. We present the methodology in the context of the Abstract DPLL Modulo Theories framework. Besides adapting many of Simplify's techniques, we also introduce a number of new heuristics. Most important is the notion of instantiation level which provides an effective mechanism for prioritizing and managing the large search space inherent in quantifier instantiation techniques. These techniques have been implemented in the SMT system CVC3. Experimental results show that our methodology enables CVC3 to solve a significant number of quantified benchmarks that were not solvable with previous approaches.

Semantic patterns for user‐interactive question answering

AbstractA new type of semantic pattern is proposed in this paper, which can be used by users to post questions and answers in user‐interactive question answering (QA) systems. The necessary procedures of using semantic patterns in a QA system are also presented, which include question structure analysis, pattern matching, pattern generation, pattern classification and answer extraction. Both the manual creation method and the automatic generation method are proposed for patterns for different applications. A pattern instantiation level metrics is also presented for the predication of the quality of generated or learned patterns. We implemented a user interface for using the semantic pattern in our QA system, which allows users to effectively post and answer questions. Copyright © 2007 John Wiley & Sons, Ltd.

Modelbase Construction with Object‐Oriented Constructs*

ABSTRACTThis paper proposes modelbase construction mechanisms in a decision support system (DSS) for representing and managing models of diverse management science/operations research modeling paradigms, using object‐oriented database management systems (ODBMS) constructs. It focuses on the construction of a modelbase that maintains logical independence among the DSS components including modelbase, database, and solvers, but relieves the mismatching characteristics by facilitating intelligent and stabilized integration of them. As a conceptual framework to build such a modelbase, this research uses generic model concepts, and adopts structured modeling language (SML) as a paradigm‐neutral model representation sublanguage. In the modelbase, three model abstraction layers including model type, model structure, and model instance are devised to facilitate the capture of multiple modeling paradigms and specific application models in different instantiation levels. The constructs and methods discussed are flexible enough to be applied to a wide variety of decision‐making and problem‐solving paradigms. A prototype system is developed under a commercial ODBMS called OBJECTSTORE with the C++ programming language, and diverse model manipulation commands are illustrated by an object‐oriented structured query language (OSQL).

Model Libraries: Knowledge Representation and Reasoning

We propose a knowledge representation and operators for large, diverse model libraries. The knowledge representation distinguishes between model types that are classes of models defined by a collection of assumptions and model templates and instances that instantiate model types through decomposition and specialization of components. Underlying the knowledge representation are inheritance to represent the level of generalization and multiple levels of instantiation to represent the level of details among models. For inheritance, we define partial order relations for models and their components including domains, classes, units, assumptions, metrics, and constraints. For instantiation, we define constraints on transformations between model types, templates, and instances. The inexact search operators support content-based retrieval especially when it is difficult for a user to exactly specify the characteristics of models of interest. The operators use similarity functions to measure the closeness of a match between a search atom and the value of a model attribute. The explanation operators are designed to enhance users' understanding about the symbols, assumptions, components, and relationships among models. A prototype system has been implemented in Prolog to demonstrate the essential features of the model library system. INFORMS Journal on Computing, ISSN 1091-9856, was published as ORSA Journal on Computing from 1989 to 1995 under ISSN 0899-1499.