Results 1 - 10
of
21
Dynamic Bayesian Networks for Information Fusion with Applications to Human-Computer Interfaces
, 1999
"... Recent advances in various display and virtual technologies coupled with an explosion in available computing power have given rise to a numberofnovel human-computer interaction (HCI) modalities -- speech, vision-based gesture recognition, eye tracking, EEG, etc. However, despite the abundance of nov ..."
Abstract
-
Cited by 37 (1 self)
- Add to MetaCart
Recent advances in various display and virtual technologies coupled with an explosion in available computing power have given rise to a numberofnovel human-computer interaction (HCI) modalities -- speech, vision-based gesture recognition, eye tracking, EEG, etc. However, despite the abundance of novel interaction devices, the naturalness and efficiency of HCI has remained low. This is due in particular to the lack of robust sensory data interpretation techniques. To deal with the task of interpreting single and multiple interaction modalities this dissertation establishes a novel probabilistic approach based on dynamic Bayesian networks (DBNs). As a generalization of the successful hidden Markov models, DBNs are a natural basis for the general temporal action interpretation task. The problem of interpretation of single or multiple interacting modalities can then be viewed as a Bayesian inference task. In this work three complex DBN models are introduced: mixtures of DBNs, mixed-state DBNs, and coupled HMMs. In-depth study of these models yields efficient approximate inference and parameter learning techniques applicable to a wide variety of problems. Experimental validation of the proposed approaches in the domains of gesture and speech recognition con rms the model's applicability to both unimodal and multimodal interpretation tasks.
Integrating Probabilistic Reasoning into Plan Recognition
- PROCEEDINGS OF THE 11TH EUROPEAN CONFERENCE ON ARTIFICIAL INTELLIGENCE
, 1994
"... Plan recognition is an important task whenever a system has to take into account an agent's actions and goals in order to be able to react adequately. Many plan recognizers, however, are only capable of deriving a set of equally plausible plan hypotheses. This is of little use whenever the syst ..."
Abstract
-
Cited by 14 (2 self)
- Add to MetaCart
(Show Context)
Plan recognition is an important task whenever a system has to take into account an agent's actions and goals in order to be able to react adequately. Many plan recognizers, however, are only capable of deriving a set of equally plausible plan hypotheses. This is of little use whenever the system actually has to react, e.g., when an intelligent help system is asked to give advice to the user of an application system or when the user is to be offered semantic plan completion. In such cases it is crucial to have a way of measuring which allows assessment of the various hypotheses and eventual selection of the "best" one if necessary. In this paper we describe how such a measure can be defined on the basis of Dempster-Shafer Theory and how it can be applied to any plan recognizer.
Valuation of permanent transitory and price irrelevant components of reported earnings
- Journal of Accounting, Auditing and Finance, V
, 1998
"... Both the economic nature of events and extant accounting rules cause reported earnings to have different components, each with different val-uation implications. The price-earnings link is described better by sepa-rating components of unexpected earnings and multiplying each by a different response ..."
Abstract
-
Cited by 11 (1 self)
- Add to MetaCart
Both the economic nature of events and extant accounting rules cause reported earnings to have different components, each with different val-uation implications. The price-earnings link is described better by sepa-rating components of unexpected earnings and multiplying each by a different response coefficient, rather than applying a single earnings re-sponse coefficient (ERC) to aggregate unexpected earnings. Using a simple model that assumes three types of innovations to reported earnings (per-manent, transitory, and price-irrelevant), we develop systematic links among current earnings components, future earnings, and stock prices. Empirical tests of the model's predictions confirm the validity of our char-acterization of the price-earnings link. Attempts to understand better the effects of growth and (beta) risk result in little improvement. 1.
On Combining Classifier Mass Functions for Text Categorization
- IEEE Transactions on Knowledge and Data Engineering
, 2005
"... Abstract—Experience shows that different text classification methods can give different results. We look here at a way of combining the results of two or more different classification methods using an evidential approach. The specific methods we have been experimenting with in our group include the ..."
Abstract
-
Cited by 10 (1 self)
- Add to MetaCart
(Show Context)
Abstract—Experience shows that different text classification methods can give different results. We look here at a way of combining the results of two or more different classification methods using an evidential approach. The specific methods we have been experimenting with in our group include the Support Vector Machine, kNN (nearest neighbors), kNN model-based approach (kNNM), and Rocchio methods, but the analysis and methods apply to any methods. We review these learning methods briefly, and then we describe our method for combining the classifiers. In a previous study, we suggested that the combination could be done using evidential operations [1] and that using only two focal points in the mass functions (see below) gives good results. However, there are conditions under which we should choose to use more focal points. We assess some aspects of this choice from an evidential reasoning perspective and suggest a refinement of the approach. Index Terms—Data mining systems and tools, modeling of structured, textual and multimedia data, uncertainty reasoning. æ 1
The Role of Domain Knowledge in Data Mining
, 1995
"... The ideal situation for a Data Mining or Knowledge Discovery system would be for the user to be able to pose a query of the form "Give me something interesting that could be useful" and for the system to discover some useful knowledge for the user. But such a system would be unrealistic as ..."
Abstract
-
Cited by 8 (0 self)
- Add to MetaCart
The ideal situation for a Data Mining or Knowledge Discovery system would be for the user to be able to pose a query of the form "Give me something interesting that could be useful" and for the system to discover some useful knowledge for the user. But such a system would be unrealistic as databases in the real world are very large and so it would be too inefficient to be workable. So the role of the human within the discovery process is essential. Moreover, the measure of what is meant by "interesting to the user" is dependent on the user as well as the domain within which the Data Mining system is being used. In this paper we discuss the use of domain knowledge within Data Mining. We define three classes of domain knowledge: Hierarchical Generalization Trees (HG-Trees), Attribute Relationship Rules (AR-rules) and EnvironmentBased Constraints (EBC). We discuss how each one of these types of domain knowledge is incorporated into the discovery process within the EDM (Evidential Data Min...
Data Fusion for Traffic Incident Detection Using D-S Evidence Theory with Probabilistic SVMs
- ISBN: 978-988-17012-0-6 ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online) WCECS 2010
"... Abstract—Accurate Incident detection is one of the important components in Intelligent Transportation Systems. It identifies traffic abnormality based on input signals obtained from different type traffic flow sensors. To date, the development of Intelligent Transportation Systems has urged the rese ..."
Abstract
-
Cited by 3 (0 self)
- Add to MetaCart
Abstract—Accurate Incident detection is one of the important components in Intelligent Transportation Systems. It identifies traffic abnormality based on input signals obtained from different type traffic flow sensors. To date, the development of Intelligent Transportation Systems has urged the researchers in incident detection area to explore new techniques with high adaptability to changing site traffic characteristics. From the viewpoint of evidence theory, information obtained from each sensor can be considered as a piece of evidence, and as such, multisensor based traffic incident detector can be viewed as a problem of evidence fusion. This paper proposes a new technique for traffic incident detection, which combines multiple multi-class probability support vector machines (MPSVM) using D-S evidence theory. We present a preliminary review of evidence theory and explain how the multi-sensor traffic incident detector problem can be framed in the context of this theory, in terms of incidents frame of discernment, mass functions is designed by mapping the outputs of standard support vector machines into a posterior probability using a learned sigmoid function. The experiment results suggest that MPSVM is a better adaptive classifier for incident detection problem with a changing site traffic environment. Index Terms—traffic incident detector, evidence theory, support vector machine, data fusion, pattern recognition I.
Cognitive Informatics: Exploring the Theoretical Foundations for Natural Intelligence, Neural Informatics, Autonomic Computing, and Agent Systems ABSTRACT Editorial PrEfacE
"... Cognitive informatics (CI) is a new discipline that studies the natural intelligence and internal information processing mechanisms of the brain, as well as the processes involved in perception and cognition. CI provides a coherent set of fundamental theories, and contemporary mathematics, ..."
Abstract
-
Cited by 3 (0 self)
- Add to MetaCart
(Show Context)
Cognitive informatics (CI) is a new discipline that studies the natural intelligence and internal information processing mechanisms of the brain, as well as the processes involved in perception and cognition. CI provides a coherent set of fundamental theories, and contemporary mathematics,
Data Mining in Parallel
- Proc. World Occam and Transputer User Group Conf
, 1995
"... . In this paper we discuss the efficient implementation of the STRIP (STrong Rule Induction in Parallel) algorithm in parallel using a transputer network. Strong rules are rules that are almost always correct. We show that STRIP is well suited for parallel implementation with scope for parallelism e ..."
Abstract
-
Cited by 2 (0 self)
- Add to MetaCart
(Show Context)
. In this paper we discuss the efficient implementation of the STRIP (STrong Rule Induction in Parallel) algorithm in parallel using a transputer network. Strong rules are rules that are almost always correct. We show that STRIP is well suited for parallel implementation with scope for parallelism existing at four different levels of the algorithm. We present a performance study analysing the best topologies for the transputer network using different number of transputers. The choice of certain variables (the number and size of samples) in the STRIP algorithm affects the performance (speedup and efficiency) of the implementation. 1. Introduction Since 1970 when Codd introduced the relational model for databases [7], the database industry has matured a great deal and applications that were never envisaged earlier have become possible. Furthermore, heterogeneous data collections, perhaps distributed and multi-media, can now be integrated and used globally [6]. Despite these and other adv...
Aspects of Uncertainty Handling for Knowledge Discovery in Databases
- University of Ulster
, 1994
"... In this paper we discuss the role of uncertainty in Knowledge Discovery in Databases (KDD) and discuss the applicability of Evidence Theory towards achieving the goal of handling the uncertainty successfully, incorporating it into the discovery process. We claim that Evidence Theory is more su ..."
Abstract
-
Cited by 1 (1 self)
- Add to MetaCart
In this paper we discuss the role of uncertainty in Knowledge Discovery in Databases (KDD) and discuss the applicability of Evidence Theory towards achieving the goal of handling the uncertainty successfully, incorporating it into the discovery process. We claim that Evidence Theory is more suitable for representing and handling uncertainty within KDD than the Bayesian Model and present a case for the same. We discuss , EDM, our framework for KDD based on Evidence Theory. EDM consists of representation methods for data and knowledge and operators on the data and knowledge that together form the discovery process. Of the different types of operators within EDM, in this paper we limit our discussion to combination operators. We introduce a combination operator called the Proportional Belief Transfer operator and discuss its properties. In particular, we show how it differs from the Dempster-Shafer Orthogonal Sum. 1. Introduction Uncertainty Handling has been an impo...
