The Experts below are selected from a list of 96 Experts worldwide ranked by ideXlab platform
Kapil Vaswani - One of the best experts on this subject based on the ideXlab platform.
-
Mining quantified temporal Rules: Formalism, algorithms, and evaluation
Science of Computer Programming, 2012Co-Authors: Ganesan Ramalingam, Venkatesh-prasad Ranganath, Kapil VaswaniAbstract:Libraries usually impose constraints on how clients should use them. Often these constraints are not well-documented. In this paper, we address the problem of recovering such constraints automatically, a problem referred to as specification mining. Given some client programs that use a given library, we identify constraints on the library usage that are (almost) satisfied by the given set of clients. The class of Rules we target for mining combines simple binary temporal operators with state predicates (composed of equality constraints) and quantification. This is a simple yet expressive subclass of temporal properties (LTL formulae) that allows us to capture many common API usage Rules. We focus on recovering Rules from execution traces and apply classical data mining concepts to be robust against bugs (API usage Rule violations) in clients. We present new algorithms for mining Rules from execution traces. We show how a Propositional Rule mining algorithm can be generalized to treat quantification and state predicates in a unified way. Our approach enables the miner to be complete (i.e. , mine all Rules within the targeted class that are satisfied by the given traces) while avoiding an exponential blowup. We have implemented these algorithms and used them to mine API usage Rules for several Windows APIs. Our experiments show the efficiency and effectiveness of our approach.
-
WCRE - Mining Quantified Temporal Rules: Formalism, Algorithms, and Evaluation
2009 16th Working Conference on Reverse Engineering, 2009Co-Authors: Ganesan Ramalingam, Venkatesh-prasad Ranganath, Kapil VaswaniAbstract:Libraries usually impose constraints on how clients should use them. Often these constraints are not well-documented. In this paper, we address the problem of recovering such constraints automatically, a problem referred to as specification mining. Given some client programs that use a given library, we identify constraints on the library usage that are (almost) satisfied by the given set of clients.The class of Rules we target for mining combines simple binary temporal operators with state predicates (involving equality constraints) and quantification. This is a simple yet expressive subclass of temporal properties that allows us to capture many common API usage Rules. We focus on recovering Rules from execution traces and apply classical data mining concepts to be robust against bugs (API usage Rule violations) in clients. We present new algorithms for mining Rules from execution traces. We show how a Propositional Rule mining algorithm can be generalized to treat quantification and state predicates in a unified way. Our approach enables the miner to be complete — mine all Rules within the targeted class that are satisfied by the given traces — while avoiding an exponential blowup.We have implemented these algorithms and used them to mine API usage Rules for several Windows APIs. Our experiments show the efficiency and effectiveness of our approach.
Jean-daniel Zucker - One of the best experts on this subject based on the ideXlab platform.
-
A framework for learning Multiple-Instance Decision Trees and Rule Sets
2001Co-Authors: Yann Chevaleyre, Jean-daniel ZuckerAbstract:This paper proposes a generic extension to Propositional Rule learners to handle multiple-instance data. In a multiple-instance representation, each learning example is represented by a bag” of fixed-length feature vectors”.Such a representation,lying somewhere between Propositional and first-order representation, offers a tradeoff between the two. Naive-RipperMi is one implementation of this extension on the Rule learning algorithm Ripper. Several pitfalls encountered by this naive extension during induction are explained. A new multiple-instance search bias based on decision tree techniques is then used to avoid these pitfalls. Experimental results show the benefits of this approach for solving Propositionalized relational problems in terms of speed and accuracy.
-
ECML - A framework for learning Rules from multiple instance data
Machine Learning: ECML 2001, 2001Co-Authors: Yann Chevaleyre, Jean-daniel ZuckerAbstract:This paper proposes a generic extension to Propositional Rule learners to handle multiple-instance data. In a multiple-instance representation, each learning example is represented by a "bag" of fixed-length "feature vectors". Such a representation, lying somewhere between Propositional and first-order representation, offers a tradeoff between the two. NAIVE-RIPPERMI is one implementation of this extension on the Rule learning algorithm RIPPER. Several pitfalls encountered by this naive extension during induction are explained. A new multiple-instance search bias based on decision tree techniques is then used to avoid these pitfalls. Experimental results show the benefits of this approach for solving Propositionalized relational problems in terms of speed and accuracy.
-
Learning Rules from Multiple Instance Data: Issues and Algorithms
2001Co-Authors: Yann Chevaleyre, Jean-daniel ZuckerAbstract:In a multiple-instance representation, each learning example is represented by a “bag” of fixed-length “feature vectors”. Such a representation, lying somewhere between Propositional and first-order representation, offers a tradeoff between the two. This paper proposes a generic extension to Propositional Rule learners to handle multiple-instance data. It describes NAIVE-RIPPERMI, an implementation of this extension on the Rule learning algorithm RIPPER. It then explains several pitfalls encountered by this naive extension during induction. It goes on to describe algorithmic modifications and a new multipleinstance coverage measure which are shown to avoid these pitfalls. Experimental results show the benefits of this approach for solving Propositionalized relational problems in terms of speed and accuracy. keywords: Multiple-instance learning problem, Rule learning, Propositionalization, relational learning, mutagenesis learning task
Johannes Furnkranz - One of the best experts on this subject based on the ideXlab platform.
-
handling unknown and imprecise attribute values in Propositional Rule learning a feature based approach
Pacific Rim International Conference on Artificial Intelligence, 2008Co-Authors: Dragan Gamberger, Nada Lavrac, Johannes FurnkranzAbstract:Rule learning systems use features as the main building blocks for Rules. A feature can be a simple attribute-value test or a test of the validity of a complex domain knowledge relationship. Most existing concept learning systems generate features in the Rule construction process. However, the separation of feature generation and Rule construction processes has several theoretical and practical advantages. In particular, the proposed transformation from the attribute to the feature space motivates a novel, theoretically justified procedure for handling of unknown attribute values. This approach suggests also a novel procedure for handling imprecision of numerical attributes. The possibility of controlling the expected imprecision of numerical attributes during the induction process is a novel machine learning concept which has a high application potential for solving real world problems.
-
PRICAI - Handling Unknown and Imprecise Attribute Values in Propositional Rule Learning: A Feature-Based Approach
PRICAI 2008: Trends in Artificial Intelligence, 2008Co-Authors: Dragan Gamberger, Nada Lavrac, Johannes FurnkranzAbstract:Rule learning systems use features as the main building blocks for Rules. A feature can be a simple attribute-value test or a test of the validity of a complex domain knowledge relationship. Most existing concept learning systems generate features in the Rule construction process. However, the separation of feature generation and Rule construction processes has several theoretical and practical advantages. In particular, the proposed transformation from the attribute to the feature space motivates a novel, theoretically justified procedure for handling of unknown attribute values. This approach suggests also a novel procedure for handling imprecision of numerical attributes. The possibility of controlling the expected imprecision of numerical attributes during the induction process is a novel machine learning concept which has a high application potential for solving real world problems.
Oriane Mattetailliez - One of the best experts on this subject based on the ideXlab platform.
-
inductive improvement of part of speech tagging and its effect on a terminology of molecular biology
Lecture Notes in Computer Science, 2005Co-Authors: Ahmed Amrani, Mathieu Roche, Yves Kodratoff, Oriane MattetailliezAbstract:In the context of Part-of-Speech (PoS)-tagging of specialized corpora, we proposed an inductive approach focusing on the most ‘important' PoStags because mistaking them can lead to a total misunderstanding of the text After a standard tagging of a biological corpus by Brill's tagger, we noted persistent errors that are very hard to deal with As an application, we studied two cases of different nature: first, confusion between past participle, adjective and preterit for verbs that end with ‘ed'; second, confusion between plural nouns and verbs, 3rd person singular present With a friendly user interface, the expert corrected the examples Then, from these well-annotated examples, we induced Rules using a Propositional Rule induction algorithm Experimental validation showed improvement in tagging precision The relevance of the terminology of the considered field, here molecular biology, is greatly improved when the number of these tagging errors decreases.
Ganesan Ramalingam - One of the best experts on this subject based on the ideXlab platform.
-
Mining quantified temporal Rules: Formalism, algorithms, and evaluation
Science of Computer Programming, 2012Co-Authors: Ganesan Ramalingam, Venkatesh-prasad Ranganath, Kapil VaswaniAbstract:Libraries usually impose constraints on how clients should use them. Often these constraints are not well-documented. In this paper, we address the problem of recovering such constraints automatically, a problem referred to as specification mining. Given some client programs that use a given library, we identify constraints on the library usage that are (almost) satisfied by the given set of clients. The class of Rules we target for mining combines simple binary temporal operators with state predicates (composed of equality constraints) and quantification. This is a simple yet expressive subclass of temporal properties (LTL formulae) that allows us to capture many common API usage Rules. We focus on recovering Rules from execution traces and apply classical data mining concepts to be robust against bugs (API usage Rule violations) in clients. We present new algorithms for mining Rules from execution traces. We show how a Propositional Rule mining algorithm can be generalized to treat quantification and state predicates in a unified way. Our approach enables the miner to be complete (i.e. , mine all Rules within the targeted class that are satisfied by the given traces) while avoiding an exponential blowup. We have implemented these algorithms and used them to mine API usage Rules for several Windows APIs. Our experiments show the efficiency and effectiveness of our approach.
-
WCRE - Mining Quantified Temporal Rules: Formalism, Algorithms, and Evaluation
2009 16th Working Conference on Reverse Engineering, 2009Co-Authors: Ganesan Ramalingam, Venkatesh-prasad Ranganath, Kapil VaswaniAbstract:Libraries usually impose constraints on how clients should use them. Often these constraints are not well-documented. In this paper, we address the problem of recovering such constraints automatically, a problem referred to as specification mining. Given some client programs that use a given library, we identify constraints on the library usage that are (almost) satisfied by the given set of clients.The class of Rules we target for mining combines simple binary temporal operators with state predicates (involving equality constraints) and quantification. This is a simple yet expressive subclass of temporal properties that allows us to capture many common API usage Rules. We focus on recovering Rules from execution traces and apply classical data mining concepts to be robust against bugs (API usage Rule violations) in clients. We present new algorithms for mining Rules from execution traces. We show how a Propositional Rule mining algorithm can be generalized to treat quantification and state predicates in a unified way. Our approach enables the miner to be complete — mine all Rules within the targeted class that are satisfied by the given traces — while avoiding an exponential blowup.We have implemented these algorithms and used them to mine API usage Rules for several Windows APIs. Our experiments show the efficiency and effectiveness of our approach.
