Results 1  10
of
21
A Maximum Entropy approach to Natural Language Processing
 COMPUTATIONAL LINGUISTICS
, 1996
"... The concept of maximum entropy can be traced back along multiple threads to Biblical times. Only recently, however, have computers become powerful enough to permit the widescale application of this concept to real world problems in statistical estimation and pattern recognition. In this paper we des ..."
Abstract

Cited by 1082 (5 self)
 Add to MetaCart
The concept of maximum entropy can be traced back along multiple threads to Biblical times. Only recently, however, have computers become powerful enough to permit the widescale application of this concept to real world problems in statistical estimation and pattern recognition. In this paper we describe a method for statistical modeling based on maximum entropy. We present a maximumlikelihood approach for automatically constructing maximum entropy models and describe how to implement this approach efficiently, using as examples several problems in natural language processing.
Towards correcting input data errors probabilistically using integrity constraints
 In MobiDB
, 2006
"... Mobile and pervasive applications frequently rely on devices such as RFID antennas or sensors (light, temperature, motion) to provide them information about the physical world. These devices, however, are unreliable. They produce streams of information where portions of data may be missing, duplicat ..."
Abstract

Cited by 42 (9 self)
 Add to MetaCart
Mobile and pervasive applications frequently rely on devices such as RFID antennas or sensors (light, temperature, motion) to provide them information about the physical world. These devices, however, are unreliable. They produce streams of information where portions of data may be missing, duplicated, or erroneous. Current state of the art is to correct errors locally (e.g., range constraints for temperature readings) or use spatial/temporal correlations (e.g., smoothing temperature readings). However, errors are often apparent only in a global setting, e.g., missed readings of objects that are known to be present, or exit readings from a parking garage without matching entry readings. In this paper, we present StreamClean, a system for correcting input data errors automatically using application defined global integrity constraints. Because it is frequently impossible to make corrections with certainty, we propose a probabilistic approach, where the system assigns to each input tuple the probability that it is correct. We show that StreamClean handles a large class of input data errors, and corrects them sufficiently fast to keepup with input rates of many mobile and pervasive applications. We also show that the probabilities assigned by StreamClean correspond to a user’s intuitive notion of correctness.
UserAdaptive Exploration of Multidimensional Data
, 2000
"... In this paper we present a tool for enhanced exploration of OLAP data that is adaptive to a user's prior knowledge of the data. The tool continuously keeps track of the parts of the cube that a user has visited. The information in these scattered visited parts of the cube is pieced together to form ..."
Abstract

Cited by 24 (1 self)
 Add to MetaCart
In this paper we present a tool for enhanced exploration of OLAP data that is adaptive to a user's prior knowledge of the data. The tool continuously keeps track of the parts of the cube that a user has visited. The information in these scattered visited parts of the cube is pieced together to form a model of the user's expected values in the unvisited parts. The mathematical foundation for this modeling is provided by the classical Maximum Entropy principle. At any time, the user can query for the most surprising unvisited parts of the cube. The most surprising values are defined as those which if known to the user would bring the new expected values closest to the actual values. This process of updating the user's context based on visited parts and querying for regions to explore further continues in a loop until the user's mental model perfectly matches the actual cube. We believe and prove through experiments that such a userintheloop exploration will enable much faster assimilation of all significant information in the data compared to existing manual explorations.
Wrapped progressive sampling search for optimizing learning algorithm parameters
 Proceedings of the Sixteenth BelgianDutch Conference on Artificial Intelligence
, 2004
"... We present a heuristic metalearning search method for finding a set of optimized algorithmic parameters for a range of machine learning algorithms. The method, wrapped progressive sampling, is a combination of classifier wrapping and progressive sampling of training data. A series of experiments on ..."
Abstract

Cited by 17 (8 self)
 Add to MetaCart
We present a heuristic metalearning search method for finding a set of optimized algorithmic parameters for a range of machine learning algorithms. The method, wrapped progressive sampling, is a combination of classifier wrapping and progressive sampling of training data. A series of experiments on UCI benchmark data sets with nominal features, and five machine learning algorithms to which simple wrapping and wrapped progressive sampling is applied, yields results that show little improvement for the algorithm which offers few parameter variations, but marked improvements for the algorithms offering many possible testable parameter combinations, yielding up to 32.2 % error reduction with the winnow learning algorithm. 1
An estimation of distribution algorithm based on maximum entropy
 GECCO 2004: Genetic and Evolutionary Computation Conference
, 2004
"... Abstract. Estimation of distribution algorithms (EDA) are similar to genetic algorithms except that they replace crossover and mutation with sampling from an estimated probability distribution. We develop a framework for estimation of distribution algorithms based on the principle of maximum entropy ..."
Abstract

Cited by 9 (0 self)
 Add to MetaCart
Abstract. Estimation of distribution algorithms (EDA) are similar to genetic algorithms except that they replace crossover and mutation with sampling from an estimated probability distribution. We develop a framework for estimation of distribution algorithms based on the principle of maximum entropy and the conservation of schema frequencies. An algorithm of this type gives better performance than a standard genetic algorithm (GA) on a number of standard test problems involving deception and epistasis (i.e. Trap and NK). 1
Ising Quantum Chain and Sequence Evolution
 J. Stat. Phys
, 1999
"... A sequence space model which describes the interplay of mutation and selection in molecular evolution is shown to be equivalent to an Ising quantum chain. Observable quantities, tailored to match the biological situation, will then be employed to treat three tness landscapes exactly. appeared in: J ..."
Abstract

Cited by 7 (5 self)
 Add to MetaCart
A sequence space model which describes the interplay of mutation and selection in molecular evolution is shown to be equivalent to an Ising quantum chain. Observable quantities, tailored to match the biological situation, will then be employed to treat three tness landscapes exactly. appeared in: J. Stat. Phys. 92 (1998), 10171052. Keywords: Biological evolution; sequence space; mutation and selection; Ising quantum chain; meaneld model; phase transition 1 Introduction Sequence space models seek to describe biological evolution at the molecular level through mutation and selection. Wellknown ones are Kauman's adaptive walk [27] and Eigen's quasispecies model [14]. Whereas the former describes a hillclimbing process of a genetically homogeneous population in tunably rugged tness landscapes (where the tness values are considered as a mountain range over sequence space), the latter includes the genetic structure of the population due to the balance between mutation and selectio...
Improving sequence segmentation learning by predicting trigrams
 In Proceedings of the Ninth Conference on Natural Language Learning, CoNLL2005 (pp. 80–87), Ann Arbor
, 2005
"... Symbolic machinelearning classifiers are known to suffer from nearsightedness when performing sequence segmentation (chunking) tasks in natural language processing: without special architectural additions ..."
Abstract

Cited by 7 (6 self)
 Add to MetaCart
Symbolic machinelearning classifiers are known to suffer from nearsightedness when performing sequence segmentation (chunking) tasks in natural language processing: without special architectural additions
W.: Evaluating hybrid versus datadriven coreference resolution
 In: Anaphora: Analysis, Algorithms and Applications (LNAI
"... resolution ..."
A Hybrid MaxEnt/HMM based ASR System
, 2005
"... The aim of this work is to develop a practical framework, which extends the classical Hidden Markov Models (HMM) for continuous speech recognition based on the Maximum Entropy (MaxEnt) principle. The MaxEnt models can estimate the posterior probabilities directly as with Hybrid NN/HMM connectionist ..."
Abstract

Cited by 2 (1 self)
 Add to MetaCart
The aim of this work is to develop a practical framework, which extends the classical Hidden Markov Models (HMM) for continuous speech recognition based on the Maximum Entropy (MaxEnt) principle. The MaxEnt models can estimate the posterior probabilities directly as with Hybrid NN/HMM connectionist speech recognition systems. In particular, a new acoustic modelling based on discriminative MaxEnt models is formulated and is being developed to replace the generative Gaussian Mixture Models (GMM) commonly used to model acoustic variability. Initial experimental results using the TIMIT phone task are reported.