In applications such as character recognition, some classes are heavily overlapped but are not necessarily to be separated. For classification of such overlapping classes, either discriminating between them or merging...
详细信息
ISBN:
(纸本)9783540699385
In applications such as character recognition, some classes are heavily overlapped but are not necessarily to be separated. For classification of such overlapping classes, either discriminating between them or merging them into a metaclass does not satisfy. Merging the overlapping classes into a metaclass implies that with in- metaclass substitution is considered as correct classification. For such classification problems, I propose a partial discriminative training (PDT) scheme for neuralnetworks, in which, a training pattern of an overlapping class is used as a positive sample of its labeled class, and neither positive nor negative sample for its allied classes (classes overlapping with the labeled class). In experiments of handwritten letter recognition using neuralnetworks and support vector machines, the PDT scheme mostly outperforms crosstraining (a scheme for multi-labeled classification), ordinary discriminative training and metaclass classification.
This paper investigates the use of artificialneuralnetworks (ANN) to mine and predict;patterns in software aging phenomenon. We analyze resource usage data collected on a typical long-running software system: a web ...
详细信息
ISBN:
(纸本)9783540699385
This paper investigates the use of artificialneuralnetworks (ANN) to mine and predict;patterns in software aging phenomenon. We analyze resource usage data collected on a typical long-running software system: a web server. A Multi-Layer Perceptron feed forwardartificialneural Network was trained on an Apache web server dataset to predict future server swap space and physical free memory resource exhaustion through ANN univariate tirne series forecasting and ANN nonlinear multivari ' ate time series empirical modeling. The results were benchmarked against those obtained from non-parametric statistical techniques, parametric time series models and other empirical modeling techniques reported in the literature.
Visual database engines are usually based on predefined criteria for retrieving the images in response to a given query. In this paper, we propose a novel approach based on neuralnetworks by which the retrieval crite...
详细信息
Visual database engines are usually based on predefined criteria for retrieving the images in response to a given query. In this paper, we propose a novel approach based on neuralnetworks by which the retrieval criterion is derived on the basis of learning from examples. In particular, the proposed approach uses a graph-based image representation that denotes the relationships among regions in the image and on recursive neuralnetworks which can process directed ordered acyclic graphs. The graph-based representation combines structural and subsymbolic features of the image, while recursive neuralnetworks can discover the optimal representation for searching the image database. A set of preliminary experiments on artificial images clearly indicate that the proposed approach is very promising. (C) 2002 Elsevier Science B.V. All rights reserved.
This paper presents a novel neural network model, called similarity neural network (SNN), designed to learn similarity measures for pairs of patterns. The model guarantees to compute a non negative and symmetric measu...
详细信息
ISBN:
(纸本)9783540699385
This paper presents a novel neural network model, called similarity neural network (SNN), designed to learn similarity measures for pairs of patterns. The model guarantees to compute a non negative and symmetric measure, and shows good generalization capabilities even if a very small set of supervised examples is used for training. Preliminary experiments, carried out on some UCI datasets, are presented, showing promising results.
In this paper, a methodology for the generation of benchmarks in patternrecognition is described. The patterns are represented by means of an attributed plex language, which are based on plex grammars augmented by at...
详细信息
In this paper, a methodology for the generation of benchmarks in patternrecognition is described. The patterns are represented by means of an attributed plex language, which are based on plex grammars augmented by attributes. It is shown that the generated patterns are particularly suitable for the extraction of graph-based representations. As a result, databases of artificial pictures and correspondent graphs can be generated. These collections of graphs are very appropriate for benchmarks in the area of structural patternrecognition, since they are originated from a grammar and not from random distributions. The tools for creating the databases are public domain and have been already used for benchmarking artificialneuralnetworks operating on structured domains. (C) 2002 Elsevier Science B.V. All rights reserved.
We have implemented a speech command system which can understand simple command sentences like "Bot lift ball" or "Bot go table" using hidden Markov models (HMMs) and associative memories with spar...
详细信息
ISBN:
(纸本)9783540699385
We have implemented a speech command system which can understand simple command sentences like "Bot lift ball" or "Bot go table" using hidden Markov models (HMMs) and associative memories with sparse distributed representations. The system is composed of three modules: (1) A set of HMMs is used on phoneme level to get a phonetic transcription of the spoken sentence, (2) a network of associative memories is used to determine the word belonging to the phonetic transcription and (3) a neural network is used on the sentence level to determine the meaning of the sentence. The system is also able to learn new object words during performance.
In this paper we propose two new ensemble combiners based on the Mixture of neuralnetworks model. In our experiments, we have applied two different network architectures on the methods based on the Mixture of neural ...
详细信息
ISBN:
(纸本)9783540699385
In this paper we propose two new ensemble combiners based on the Mixture of neuralnetworks model. In our experiments, we have applied two different network architectures on the methods based on the Mixture of neuralnetworks: the Basic Network (BN) and the Multilayer Feedforward Network (MF). Moreover, we have used ensembles of MF networks previously trained with Simple Ensemble to test the performance of the combiners we propose. Finally, we compare the mixture combiners proposed with three different mixture models and other traditional combiners. The results show that the mixture combiners proposed are the best way to build Multi-net systems among the methods studied in the paper in general.
To win a board-game or more generally to gain something specific in a given Markov-environment, it is most important to have a policy in cboosing and taking actions that leads to one of several qualitative good states...
详细信息
ISBN:
(纸本)9783540699385
To win a board-game or more generally to gain something specific in a given Markov-environment, it is most important to have a policy in cboosing and taking actions that leads to one of several qualitative good states. In this paper we describe a novel method to learn a game-winning strategy. The method predicts statistical probabilities to win in given game states using a state-value function that is approximated by a Multi-layer perceptron. Those predictions will improve according to rewards given in terminal states. We have deployed that method in the game Connect Four and have compared its gameperformance with Velena [5].
Prototype based classifiers so far can only work with hard labels on the training data. In order to allow for soft labels as input label and answer, we enhanced the original LVQ algorithm. The key idea is adapting the...
详细信息
ISBN:
(纸本)9783540699385
Prototype based classifiers so far can only work with hard labels on the training data. In order to allow for soft labels as input label and answer, we enhanced the original LVQ algorithm. The key idea is adapting the prototypes depending on the similarity of their fuzzy labels to the ones of training samples. In experiments, the performance of the fuzzy LVQ was compared against the original approach. Of special interest was the behaviour of the two approaches, once noise was added to the training labels, and here a clear advantage of fuzzy versus hard training labels could be shown.
This book constitutes the refereed proceedings of the 5th INNS iapr TC3 GIRPR International workshop on artificialneuralnetworks in patternrecognition, ANNPR 2012, held in Trento, Italy, in September 2012. The 21 r...
ISBN:
(数字)9783642332128
ISBN:
(纸本)9783642332111
This book constitutes the refereed proceedings of the 5th INNS iapr TC3 GIRPR International workshop on artificialneuralnetworks in patternrecognition, ANNPR 2012, held in Trento, Italy, in September 2012. The 21 revised full papers presented were carefully reviewed and selected for inclusion in this volume. They cover a large range of topics in the field of neural network- and machine learning-based patternrecognition presenting and discussing the latest research, results, and ideas in these areas.
暂无评论