Structured Hidden Markov Model versus String Kernel Machines for Symbolic Sequence Classification.

Statistical Characterization of a Computer Grid.

Abstracting Complex Interaction Networks.
The exploration of complex interaction networks has attracted considerable interest in various fields, ranging from fundamental biology and medicine to statistical physics and information technology. In “-omics” disciplines, significant progresses have been made in understanding the large-scale properties and the biological relevance of these interactions. Some properties such as “scale-free ” distribution of nodes connectivity or “centrality ” are aspects commonly described in such complex interaction systems. In many of these studies the analysis of network topology is complemented by a semantic analysis that may rely on different labels associated to the interacting entities. One of the bottleneck of these semantic analysis is that they are computationally costly. In this paper we present a framework to explore abstraction of networks useful to speedup the computation of ground network measures. Such abstraction mechanisms may be used to efficiently provide accurate approximations of ground network measures.
A Comparison between Two Statistical Relational Models.

Image Databases Browsing by Unsupervised Learning.

Structured Hidden Markov Models A General Tool for Modeling Agent Behaviors.

Structured Hidden Markov Model A General Framework for Modeling Complex Sequences.

Modeling Temporal Behavior via Structured Hidden Markov Models an Application to Keystroking Dynamics.

Incremental Construction of Structured Hidden Markov Models.

Formalizing the Abstraction Process in Model%2DBased Diagnosis.

Abstraction and Complexity Measures.
This document describes Cilk-5.1, a language for multithreaded parallel programming based on ANSI C. Cilk is designed for computations with dynamic, highly asynchronous parallelism, which can be difficult to write in data-parallel or messagepassing style. Divide-and-conquer algorithms and tree search are examples of computations that are particularly well suited to Cilk. (For example, our group has developed three world-class chess programs, ?Tech [20], ?Socrates [18], and Cilkchess.) Besides search, we have experimented with numerical algorithms such as matrix factorization [5] and N-body simulations, and we are working on other types of applications. Unlike many other multithreaded programming systems, Cilk is algorithmic, in that the runtime system employs a scheduler that allows the performance of programs to be estimated accurately [6] based on abstract complexity measures. 1.1 What is included in Cilk-5.1?
Feature construction from synergic pairs to improve microarray%2Dbased classification.

Hypothesis Diversity in Ensemble Classification.

Machine Learning and Data Mining.
this article, we focus primarily on the role of machine learning algorithms in the data mining process. The above medical example thus represents a prototypical data mining problem in which the data consists of a collection of time series descriptions, and we use the data to learn to predict later events in the series (emergency C-section) based on earlier events (symptoms before delivery). Although we used a medical example for illustration, we could have given an analogous example of learning to predict which bank loan applicants are at high risk of failing to repay their loan (see Figure 2). As shown in the figure, in this application data typically consists of time series descriptions of customers&#039; bank balances and other demographic information, rather than medical symptoms. Yet other applications, shown in Figure 3, include predicting customer purchase behavior, customer retention, and the quality of goods produced on a particular manufacturing line. All of these are applications for which data mining has been successfully applied in practice, and where further research promises yet more effective techniques. 3 The State of the Art, and What Next?
Abstraction
software developers are using the Java language as the language of choice on many applications. This is due to the effective use of the object-oriented (OO) paradigm to develop large software projects and the ability of the Java language to support the increasing use of web technologies in business applications. The recent release of the Java version 5.0 has further increased its popularity due to the inclusion of new features that exist in other OO languages. The transition from Java 1.4.x to Java 1.5.x has provided the programmer with more flexibility when implementing programs in Java. In this paper we present the first study that investigates how the characteristics of a class are combined, thereby providing feedback on how the features provided by Java 1.4.x or earlier and Java 1.5.x or earlier are currently used. The study uses a taxonomy of OO classes that provides a mechanism to catalog any class written in Java into one of a finite set of groups. A detailed description on how we enumerated all the possible groups of Java classes is also provided. Using TaxTOOLJ (a Taxonomy Tool for the Object-Oriented Language Java) we cataloged over 155k classes from a cross-section of Java applications written in Java 1.4.x and Java 1.5.x to identify the distribution of groups used by developers. We use the data from the study to create prediction models that would allow developers to estimate the number of different groups of classes, fields and methods that are expected to be generated for large Java applications. This knowledge would be of significant benefit to aid developers in testing and maintenance activities during the software process. 1

Experimental Evaluation of Hierarchical Hidden Markov Models.


Learning Complex Event Descriptions by Abstraction.

Learning Profiles Based on Hierarchical Hidden Markov Model.

Proceedings of the 16th Eureopean Conference on Artificial Intelligence

A Monte Carlo analysis of ensemble classification.
In this paper we extend previous results providing a theoretical analysis of a new Monte Carlo ensemble classifier. The framework allows us to characterize the conditions under which the ensemble approach can be expected to outperform the single hypothesis classifier. Moreover, we provide a closed form expression for the distribution of the true ensemble accuracy, as well as of its mean and variance. We then exploit this result in order to analyze the expected error behavior in a particularly interesting case. 1.


Transition de phase et apprentissage relationnel  une analyse critique.

Monte Carlo Theory as an Explanation of Bagging and Boosting.
In this paper we propose the framework of Monte Carlo algorithms as a useful one to analyze ensemble learning. In particular, this framework allows one to guess when bagging will be useful, explains why increasing the margin improves performances, and suggests a new way of performing ensemble learning and error
Relational Learning as Search in a Critical Region.
Machine learning strongly relies on the covering test to assess whether a candidate hypothesis covers  training examples. The present paper investigates learning relational concepts from examples,  termed relational learning or inductive logic programming. In particular, it investigates the chances  of success and the computational cost of relational learning, which appears to be severely affected  by the presence of a phase transition in the covering test. To this aim, three up-to-date relational  learners have been applied to a wide range of artificial, fully relational learning problems. A first  experimental observation is that the phase transition behaves as an attractor for relational learning;  no matter which region the learning problem belongs to, all three learners produce hypotheses  lying within or close to the phase transition region. Second, a failure region appears. All three  learners fail to learn any accurate hypothesis in this region. Quite surprisingly, the probability of  failure does not systematically increase with the size of the underlying target concept: under some  circumstances, longer concepts may be easier to accurately approximate than shorter ones. Some  interpretations for these findings are proposed and discussed.
Is a Greedy Covering Strategy an Extreme Boosting%3F

Abstracting Visual Percepts to Learn Concepts.
To efficiently identify properties from its environment is an essential ability of a mobile robot who needs to interact with humans. Successful approaches to provide robots with...
Boosting as a Monte Carlo Algorithm.

Learning on the Phase Transition Edge.

Integrated Architectures for Machine Learning.

A Model of Abstraction in Visual Perception.

Analyzing Relational Learning in the Phase Transition Framework.

Can Relational Learning Scale Up%3F

Perception%2DBased Granularity Levels in Concept Representation.

Abstraction in Cartographic Generalization.
. One challenge in spatial data modelling is related to multiple representation of geographic objects. While often associated with cartographic change of scale and generalization, it applies in fact to a wider range of data analysis situations. We discuss multiple representation issues from a data modelling point of view, with emphasis on hierarchical structures inherent to multiple resolution levels. Then, we consider specific features related to multiple representation from a database modelling and querying point of view. Some guidelines for extending the object-oriented query language O2SQL are finally given. Although motivated by spatial applications, this extension is useful for a variety of other applications such as cooperative query answering [CCL91]. In particular it allows querying without exact data knowledge. 1 Introduction  We consider in this paper modelling and querying issues related to multiple representation of objects within Geographic Information Systems (GIS). In t...
Abstraction and Phase Transitions in Relational Learning.

Guest Editors%27 Introduction.

Phase Transitions in Relational Learning.
Machine learning strongly relies on the covering test to assess whether a candidate hypothesis covers  training examples. The present paper investigates learning relational concepts from examples,  termed relational learning or inductive logic programming. In particular, it investigates the chances  of success and the computational cost of relational learning, which appears to be severely affected  by the presence of a phase transition in the covering test. To this aim, three up-to-date relational  learners have been applied to a wide range of artificial, fully relational learning problems. A first  experimental observation is that the phase transition behaves as an attractor for relational learning;  no matter which region the learning problem belongs to, all three learners produce hypotheses  lying within or close to the phase transition region. Second, a failure region appears. All three  learners fail to learn any accurate hypothesis in this region. Quite surprisingly, the probability of  failure does not systematically increase with the size of the underlying target concept: under some  circumstances, longer concepts may be easier to accurately approximate than shorter ones. Some  interpretations for these findings are proposed and discussed.
Cartographic Generalization as a Combination of Representing and Abstracting Knowledge.
This article shows that cartographic generalization is best viewed as representing (formulating, renaming knowledge) and abstracting (simplifying a given representation). The general process of creating map is described so as to show how it fits into an abstraction framework developed in artificial intelligence to emphasize the difference between abstraction and representation. The utility of the framework lies in its efficiency to automate knowledge acquisition for the cartographic generalization as a combined acquisition of knowledge for abstraction and knowledge for changing a representation.  Keywords  Cartographic Generalization, Abstraction, Representation.  1. INTRODUCTION  In this paper we address the problem of automating cartographic generalization. This automation is needed for several reasons: first to decrease cost and time necessary to produce maps, then to allow geography experts who are not necessary cartography specialists to create their own maps with a good quality, ...
Relational Learning Hard Problems and Phase Transitions.

An Experimental Study of Phase Transitions in Matching.

On%2DLine Estimation of Matching Complexity in First Order Logic.

Coevolutionary%2C Distributed Search for Inducing Concept Description.

An Experimental Evaluation of Coevolutive Concept Learning.
In this paper an extensive experimental evaluation of an evolutionary approach to concept learning is presented. The experimentation, performed with the system G-NET, investigates the effectiveness of the approach along the following dimensions: Robustness with respect to parameter setting, effectiveness of the MDL criterion coupled with a stochastic search bias, impact of coevolution on the quality of the solution and on the computational effort required, and ability to face problems requiring structured representation languages. A discussion of the obtained results and a suggestion on when this type of approach might be useful is also provided. 1 INTRODUCTION  Supervised concept learning has been tackled, so far, with several approaches, including symbolic, connectionist and evolutive ones. Different approaches are better suited to different classes of problems, depending, for instance, on the nature of data or the availability of domain-specific knowledge. In the hope of making a li...
Learning in the %60%60 Real World%27%27.
. In this paper we define and characterize the process of developing a &#034;real-world&#034; Machine Learning application, with its difficulties and relevant issues, distinguishing it from the popular practice of exploiting ready-to-use data sets. To this aim, we analyze and summarize the lessons learned from applying Machine Learning techniques to a variety of problems. We believe that these lessons, though primarily based on our personal experience, can be generalized to a wider range of situations and are supported by the reported experiences of other researchers.  Keywords: Real-world applications, Application life cycles 1. Introduction  Notwithstanding the large number of Machine Learning (ML) approaches and offthe -shelf systems, significant applications with a well-documented economic and technological impact are still rare. This situation is not peculiar to ML; it applies in general to the process of technology transfer, as Isaacs and Tang (1996) note, especially when the source of tra...
Modelling Conceptual Change An Interdisciplinary Approach.

A Network Genetic Algorithm for Concept Learning.
This paper presents a highly parallel genetic algorithm, designed for concept induction in propositional and first order logics. The system exploits niches and species for learning multimodal concepts; it deeply differs from other systems because of the distributed architecture, which totally eliminates the concept of common memory. A first implementation of the system, designed for checking the possibility of exploiting parallel processing in network computer, is evaluated on standard benchmarks. The experimental results show that the system reaches good performances both with respect to the quality of the learned knowledge and with respect to the speed up on a workstation cluster. 1 INTRODUCTION  In the recent literature, Genetic Algorithms (GAs) emerged as valuable search tools in the field of concept induction (De Jong et al., 1993; Janikow, 1993; Greene &amp; Smith, 1993; Giordana &amp; Neri, 1996). Specifically, two of their features look particularly attractive, namely the exploration p...
World Model Construction in Children during Physics Learning.

A Coevolutionary Approach to Concept Learning.

Machine Learning for Information Extraction.
Comparative evaluation of Machine Learning (ML) systems used for Information Extraction (IE) has suffered from various inconsistencies in experimental procedures. This paper reports on the results of the Pascal Challenge on Evaluating Machine Learning for Information Extraction, which provides a standardised corpus, set of tasks, and evaluation methodology. The challenge is described and the systems submitted by the ten participants are briefly introduced and their performance is analysed.
Integrating Multiple Learning Strategies in First Order Logics.
. This paper describes a representation framework, that offers a unifying platform for alternative systems, which learn concepts in First Order Logic. The main aspects of this framework are discussed. First of all, the separation between the hypothesis logical language (a version of the  V L21 language) and the representation of data by means of a relational database is motivated. Then, the functional layer between data and hypotheses, which makes the data accessible by the logical level through a set of abstract properties. A novelty, in the hypothesis representation language, is the introduction of the construct of internal disjunction; such a construct, first used by the AQ and Induce systems, is here made operational via a set of algorithms, capable to learn it, for both the discrete and the continuous-valued attributes case. These algorithms are embedded in learning systems (SMART+, REGAL, SNAP, WHY, RTL) using different paradigms (symbolic, genetic or connectionist), thus realizi...
A Fine%2DGrained Parallel Evolutionary Program for Concept Induction.

Representation Change in Machine Learning.

An Analysis of the %60%60Universal Suffrage%27%27 Selection Operator.
be, 1969] Watanabe, S. (1969). Knowing and Guessing - A Formal and Qualitative  Study. Wiley, New York, NY.  [Weiss and Indurkhya, 1994] Weiss, S. M. and Indurkhya, N. (1994). Small sample decision tree pruning. In 11th International Machine Learning Conference, New Brunswick, NJ.  [Wilson, 1987] Wilson, S. (1987). Classifier systems and the animat problem. Machine Learning, 2:199--228.  [Wittgenstein, 1953] Wittgenstein, L. (1953). In Ascombe, G. E. M., editor, Philosophical  Investigations. Oxford: Blackwell, London, UK.  [Wnek and Michalski, 1994] Wnek, J. and Michalski, R. (1994). Hypothesis-driven constructive induction in aq17-hci: A method and experiments. Machine Learning, 14:136--247.  BIBLIOGRAPHY 159 of the 8th International Workshop on Machine Learning, pages 203--207, Evanston, IL. Morgan Kaufmann.  [Slotta and Chi, 1996] Slotta, J. and Chi, M. T. H. (1996). Understandin
Exploring the Power of Genetic Search in Learning Symbolic Classifiers.
be, 1969] Watanabe, S. (1969). Knowing and Guessing - A Formal and Qualitative  Study. Wiley, New York, NY.  [Weiss and Indurkhya, 1994] Weiss, S. M. and Indurkhya, N. (1994). Small sample decision tree pruning. In 11th International Machine Learning Conference, New Brunswick, NJ.  [Wilson, 1987] Wilson, S. (1987). Classifier systems and the animat problem. Machine Learning, 2:199--228.  [Wittgenstein, 1953] Wittgenstein, L. (1953). In Ascombe, G. E. M., editor, Philosophical  Investigations. Oxford: Blackwell, London, UK.  [Wnek and Michalski, 1994] Wnek, J. and Michalski, R. (1994). Hypothesis-driven constructive induction in aq17-hci: A method and experiments. Machine Learning, 14:136--247.  BIBLIOGRAPHY 159 of the 8th International Workshop on Machine Learning, pages 203--207, Evanston, IL. Morgan Kaufmann.  [Slotta and Chi, 1996] Slotta, J. and Chi, M. T. H. (1996). Understandin
Analysis of Genetic Algorithms Evolution under Pure Selection.

Learning Disjunctive Concept Definitions Using a Genetic Algorithm.

Knowledge Representation in Machine Learning.
Introduction  Knowledge representation is a topic poorly discussed in machine learning. However, it is perhaps the fundamental consideration in the design of any learning system, because the representation used determines to a great degree what can and cannot be learned. This chapter presents a survey of different schemes used to represent learned knowledge in machine learning systems. Firstly, a brief description is given of the task of designing an adaptive system and of the central role which the method of representing initial and learned knowledge plays. Secondly we examine different representational schemes which have been used in learning systems, discussing the motivation for them, the way in which they are used and their limitations. The reader should note that we do not discuss representations appropriate for connectionist or genetic approaches to learning. This is not to dismiss them as less important, but merely to constrain the scope of this chapter to a tractable s
Learning Disjunctive Concepts by Means of Genetic Algorithms.
be, 1969] Watanabe, S. (1969). Knowing and Guessing - A Formal and Qualitative  Study. Wiley, New York, NY.  [Weiss and Indurkhya, 1994] Weiss, S. M. and Indurkhya, N. (1994). Small sample decision tree pruning. In 11th International Machine Learning Conference, New Brunswick, NJ.  [Wilson, 1987] Wilson, S. (1987). Classifier systems and the animat problem. Machine Learning, 2:199--228.  [Wittgenstein, 1953] Wittgenstein, L. (1953). In Ascombe, G. E. M., editor, Philosophical  Investigations. Oxford: Blackwell, London, UK.  [Wnek and Michalski, 1994] Wnek, J. and Michalski, R. (1994). Hypothesis-driven constructive induction in aq17-hci: A method and experiments. Machine Learning, 14:136--247.  BIBLIOGRAPHY 159 of the 8th International Workshop on Machine Learning, pages 203--207, Evanston, IL. Morgan Kaufmann.  [Slotta and Chi, 1996] Slotta, J. and Chi, M. T. H. (1996). Understandin
Formal Models of Selection in Genetic Algorithms.
be, 1969] Watanabe, S. (1969). Knowing and Guessing - A Formal and Qualitative  Study. Wiley, New York, NY.  [Weiss and Indurkhya, 1994] Weiss, S. M. and Indurkhya, N. (1994). Small sample decision tree pruning. In 11th International Machine Learning Conference, New Brunswick, NJ.  [Wilson, 1987] Wilson, S. (1987). Classifier systems and the animat problem. Machine Learning, 2:199--228.  [Wittgenstein, 1953] Wittgenstein, L. (1953). In Ascombe, G. E. M., editor, Philosophical  Investigations. Oxford: Blackwell, London, UK.  [Wnek and Michalski, 1994] Wnek, J. and Michalski, R. (1994). Hypothesis-driven constructive induction in aq17-hci: A method and experiments. Machine Learning, 14:136--247.  BIBLIOGRAPHY 159 of the 8th International Workshop on Machine Learning, pages 203--207, Evanston, IL. Morgan Kaufmann.  [Slotta and Chi, 1996] Slotta, J. and Chi, M. T. H. (1996). Understandin
Learning Disjunctive Concepts with Distributed Genetic Algorithms.
this paper, REGAL [Giordana &amp; Saitta, 1993], encodes first order logic (FOL) formulas j into bit strings of fixed length. In order to make this possible, REGAL&#039;s representation language, L, has been constrained to be finite, by defining a language template, L, which represents the maximally complex formula in L. The resulting language contains FOL formulas built up with conjunction, disjunction, negation, internal disjunction and existential and universal quantification. In REGAL five genetic operators have been defined: the two-point and the uniform crossovers [De Jong, 1975; Goldberg, 1989], the generalizing and specializing crossovers, specifically designed for the task at hand [Giordana &amp; Sale, 1992], and the seeding operator, primarily used to initialize the population, in order to start with a set of formulas, each covering at least one example of the concept h [Giordana &amp; Saitta, 1993]. In the basic genetic algorithm an improved version of the sharing functions mechanism [Goldberg &amp; Richardson, 1987] has been introduced, in order to allow the formation of subpopulations. This is useful when concepts have a multimodal structure, as will be discussed in the following sections. 2. Learning Disjunctive Concepts The fundamental novelty of REGAL is its ability to learn disjunctive concepts. A disjunctive concept requires a description of the form: j 1 Ú j 2 Ú .... Ú j m ® h, (2) where each formula j j (1  j  m) covers a region of the instance space and corresponds to a different mode of the concept. If the fitness function f(j), associated to a formula j, takes into account, as usual, completeness and consistency [Michalski, 1983], then f(j) is likely to exhibit several maxima, in correspondance to the modes of the concept. + In presence of noise these strict condition...
Learning Relations Using Genetic Algorithms.
be, 1969] Watanabe, S. (1969). Knowing and Guessing - A Formal and Qualitative  Study. Wiley, New York, NY.  [Weiss and Indurkhya, 1994] Weiss, S. M. and Indurkhya, N. (1994). Small sample decision tree pruning. In 11th International Machine Learning Conference, New Brunswick, NJ.  [Wilson, 1987] Wilson, S. (1987). Classifier systems and the animat problem. Machine Learning, 2:199--228.  [Wittgenstein, 1953] Wittgenstein, L. (1953). In Ascombe, G. E. M., editor, Philosophical  Investigations. Oxford: Blackwell, London, UK.  [Wnek and Michalski, 1994] Wnek, J. and Michalski, R. (1994). Hypothesis-driven constructive induction in aq17-hci: A method and experiments. Machine Learning, 14:136--247.  BIBLIOGRAPHY 159 of the 8th International Workshop on Machine Learning, pages 203--207, Evanston, IL. Morgan Kaufmann.  [Slotta and Chi, 1996] Slotta, J. and Chi, M. T. H. (1996). Understandin
Learning Simple Recursive Theories.
Induction of recursive theories in the normal ILP setting is a difficult learning task whose  complexity is equivalent to multiple predicate learning. In this paper we propose computational  solutions to some relevant issues raised by the multiple predicate learning problem. A separate-andparallel  -conquer search strategy is adopted to interleave the learning of clauses supplying predicates  with mutually recursive definitions. A novel generality order to be imposed on the search space  of clauses is investigated, in order to cope with recursion in a more suitable way. The consistency  recovery is performed by reformulating the current theory and by applying a layering technique,  based on the collapsed dependency graph. The proposed approach has been implemented in the  ILP system ATRE and tested on some laboratory-sized and real-world data sets. Experimental results  demonstrate that ATRE is able to learn correct theories autonomously and to discover concept  dependencies. Finally, related works and their main differences with our approach are discussed.
ENIGMA A System That Learns Diagnostic Knowledge.

Pattern Recognition and Valiant%27s Learning Framework.
be, 1969] Watanabe, S. (1969). Knowing and Guessing - A Formal and Qualitative  Study. Wiley, New York, NY.  [Weiss and Indurkhya, 1994] Weiss, S. M. and Indurkhya, N. (1994). Small sample decision tree pruning. In 11th International Machine Learning Conference, New Brunswick, NJ.  [Wilson, 1987] Wilson, S. (1987). Classifier systems and the animat problem. Machine Learning, 2:199--228.  [Wittgenstein, 1953] Wittgenstein, L. (1953). In Ascombe, G. E. M., editor, Philosophical  Investigations. Oxford: Blackwell, London, UK.  [Wnek and Michalski, 1994] Wnek, J. and Michalski, R. (1994). Hypothesis-driven constructive induction in aq17-hci: A method and experiments. Machine Learning, 14:136--247.  BIBLIOGRAPHY 159 of the 8th International Workshop on Machine Learning, pages 203--207, Evanston, IL. Morgan Kaufmann.  [Slotta and Chi, 1996] Slotta, J. and Chi, M. T. H. (1996). Understandin
Multistrategy Learning and Theory Revision.
This paper presents the system WHY, which learns and updates a diagnostic knowledge base using domain knowledge and a set of examples. The a-priori knowledge consists of a causal model of the domain, stating the relationships among basic phenomena, and a body of phenomenological theory, describing the links between abstract concepts and their possible manifestations in the world. The phenomenological knowledge is used deductively, the causal model is used abductively and the examples are used inductively. The problems of imperfection and intractability of the theory are handled by allowing the system to make assumptions during its reasoning. In this way, robust knowledge can be learned with limited complexity and limited number of examples. The system works in a first order logic environment and has been applied in a real domain.  2 1. Introduction  Several authors have advocated the necessity of using deep models of the structure and behaviour of the entities involved in a given doma...
Comparison of Search Strategies in Learning Relations.

Use of Causal Models and Abduction in Learning Diagnostic Knowledge.

Learning Mutually Dependent Relations.

Use of a Causal Model to Learn Diagnostic Knowledge in a Real Domain.

Extending Inverse Resolution to Build Up Abstractions.

Abstracting Background Knowledge for Concept Learning.
ion in problem solving and learning. In Proceedings of the International Join Conference on Artifical Intelligence-89, pages 590--595, Detroit, MI, 1989. Morgan Kaufmann. [ Vorberg and Goebel, 1991 ] D. Vorberg and R. Goebel. Das losen rekursiver programmierprobleme: Rekursionsschemata. Kognitionswissenschaft, 1:83--95, 1991. [ Yoo and Fisher, 1991 ] J. Yoo and D. Fisher. Concept formation over explanations and problem-solving experience. In J. Mylopoulos and R. Reiter, editors, Proceedings of the Twelfth International Conference on Artificial Intelligence, volume 2, pages 630--637, San Mateo, CA, 1991. Morgan Kaufmann.  [ Michalski and Kodratoff, 1990 ] R. S. Michalski and Y. Kodratoff. Research in machine learning: Recent progress, classification of methods, and future directions. In Y. Kodratoff and R. S. Michalski, editors, Machine learning: An artificial intelligence approach,  volume 3, chapter 1, pages 3--30. Morgan Kaufmann, San Mateo, CA, 1990. [ Minton et al., 1989 ] S. Mint...
Panel Causality and Learning.

Investigating the Relationships between Abduction and Inverse Resolution in Proposition Calculus.

Abstracting Concepts with Inverse Resolution.

Improving Learning Using Causality and Abduction.

Integrated Learning in a Real Domain.

Rigel An Inductive Learning System.

Biasing Induction by Using a Domain Theory An Experimental Evaluation.

Integrated Learning in a real Domain.

Knowledge representation and use in pattern analysis.

Uncertainty and Intelligent Systems
A complete description of intelligent or cognitive systems always requires a model of reasoning. In real world situations reasoning is never sharp. Therefore an interval statistical formalism is presented for reasoning under conditions of uncertainty on different levels. After a short outline of the theoretical foundations, the possibilities for a representation of psychological insights are discussed. Although substantial new aspects on the relationship between logic and statistics are provided the final example and the discussion of other theories are largely tutorial. I. INTRODUCTION  A. Uncertainty  The formal semantical interpretation of a proposition may be considered as the assignment of a truth value. In case of vagueness or unfulfilled presupposition of reference, the meaning of a proposition might become indeterminate, i.e. it can neither said to be true nor false. Thus, when assigning truth values to propositions there are three possibilities: the proposition is true, false ...
Concept Acquisition in an Integrated EBL and SBL Environment.

Improving Knowledge%2DBased System Performance by Experience.

Automated Concept Acquisition in Noisy Environments.

Integrating EBL and SBL Approaches to Knowledge Base Refinement.

Knowledge base organization in expert systems.

Discrimination of Words in a Large Vocabulary Using Phonetic Descriptions.

Modelling Production Systems by means of Predicate%2DTransitions Networks.

An expert system oriented to complex pattern recognition problems.

Evidence Combination in Expert Systems.

An expert system for mapping acoustic cues into phonetic features.

Experiments in Evidence Composition in a Speech Understanding System.

An Expert System for Interpreting Speech Patterns.

An Expert System for Speech Decoding.

Automatic learning of fuzzy naming relations over finite languages.

Complexity of lambda%2Dterm reductions.

