Showing papers on "Feature (machine learning) published in 1990"

Variants of self-organizing maps

Abstract: Self-organizing maps have a bearing on traditional vector quantization. A characteristic that makes them more closely resemble certain biological brain maps, however, is the spatial order of their responses, which is formed in the learning process. A discussion is presented of the basic algorithms and two innovations: dynamic weighting of the input signals at each input of each cell, which improves the ordering when very different input signals are used, and definition of neighborhoods in the learning algorithm by the minimal spanning tree, which provides a far better and faster approximation of prominently structured density functions. It is cautioned that if the maps are used for pattern recognition and decision process, it is necessary to fine tune the reference vectors so that they directly define the decision borders. >

Statistical Pattern Integration for Mineral Exploration

Abstract: The method of statistical pattern integration used in this paper consists of reducing each set of mineral deposit indicator features on a map to a pattern of relatively few discrete states. In its simplest form the pattern for a feature is binary representing its presence or absence within a small unit cell; for example, with area of 1 km 2 on a 1:250,000 map. The feature of interest need not occur within the unit cell; its “presence” may indicate that the unit cell occurs within a given distance from a linear or curvilinear feature on a geoscience map. By using Bayes' rule, two probabilities can be computed that the unit cell contains a deposit. The log odds of the unit cell's posterior probability is obtained by adding weights W + or W − for presence or absence of the feature to the log odds of the prior probability. If a binary pattern is positively correlated with deposits, W + is positive and the contrast C=W + −W − provides a measure of the strength of this correlation. Weights for patterns with more than two states also can be computed and special consideration can be given to unknown data. Addition of weights from several patterns results in an integrated pattern of posterior probabilities. This final map subdivides the study region into areas of unit cells with different probabilities of containing a mineral deposit. In this paper, statistical pattern integration is applied to occurrence of gold mineralization in Meguma Terrane, eastern mainland Nova Scotia, Canada.

Syntactic pattern recognition of the ECG

Abstract: An application of the syntactic method to electrocardiogram (ECG) pattern recognition and parameter measurement is presented. Solutions to the related problems of primitive pattern selection, primitive pattern extraction, linguistic representation, and pattern grammar formulation are given. Attribute grammars are used as the model for the pattern grammar because of their descriptive power, founded upon their ability to handle syntactic as well as semantic information. This approach has been implemented and the performance of the resultant system has been evaluated using an annotated standard ECG library. >

Parallel self-organizing feature maps for unsupervised pattern recognition

Abstract: Neural network research has recently undergone a revival for use in pattern recognition applications.1 If a training set of data can be provided, the supervised types of networks, such as the Hopfield nets or perceptrons, can be used to recognize patterns. 10·11·18 For unsupervised pattern recognition, systems such as those of the Carpenter/Grossberg ART2 system8 and Kohonens’ self-organizing feature maps11 are the most commonly used The problem of poor separability of input vectors was recently addressed by Keller and Hunt with the fuzzy perceptron model. 13 However, with the exception of the ART2 system, none of these systems are capable of producing continuous valued output, as would be a desirable model for representation of non-distinct input vectors. This paper presents four new algorithms based on the Kohonen self-organizing feature maps which are capable of generating a continuous valued output. 4 We also present the results of some experimental studies run on the NCUBE/10 hypercube at the Univers...

A hybrid feature-based and template matching optical character recognition system

Abstract: A feature-based character recognition identification and confidence level are determined for an unknown symbol. If the confidence level is within an intermediate range, the feature-based identification is confirmed by matching the unknown character with a reference template corresponding to the feature-based identification. If the confidence level is below the intermediate range, template matching character recognition is substituted in place of the feature-based identification. If the template matching recognition identifies more than one symbol, corresponding templates from a second set of templates having thicker character strokes are employed to resolve the ambiguity.

Script Recognition with Hierarchical Feature Maps

Abstract: The hierarchical feature map system recognizes an input story as an instance of a particular script by classifying it at three levels: scripts, tracks and role bindings. The recognition taxonomy, i.e. the breakdown of each script into the tracks and roles, is extracted automatically and independently for each script from examples of script instantiations in an unsupervised self-organizing process. The process resembles human learning in that the differentiation of the most frequently encountered scripts become gradually the most detailed. The resulting structure is a hierarchical pyramid of feature maps. The hierarchy visualizes the taxonomy and the maps lay out the topology of each level. The number of input lines and the self-organization time are considerably reduced compared to the ordinary single-level feature mapping. The system can recognize incomplete stories and recover the missing events. The taxonomy also serves as memory organization for script-based episodic memory. The maps assign a unique memory location for each script instantiation. The most salient parts of the input data are separated and most resources are concentrated on representing them accurately.

Robust speaker-independent word recognition using static, dynamic and acceleration features: experiments with Lombard and noisy speech

Abstract: Speaker-independent recognition of Lombard and noisy speech by a recognizer trained with normal speech is discussed. Speech was represented by static, dynamic (first difference), and acceleration (second difference) features. Strong interaction was found between these temporal features, the frequency differentiation due to cepstral weighting, and the degree of smoothing in the spectral analysis. When combined with the other features, acceleration raised recognition rates for Lombard or noisy input speech. Dynamic and acceleration features were found to perform much better than the static feature for noisy Lombard speech. This suggests that an algorithm which excludes the static feature in high ambient noise is desirable. >

Speaker-independent word recognition using a neural prediction model

Abstract: A speech recognition model called the neural prediction model (NPM) is proposed. The model uses a sequence of multilayer perceptrons (MLPs) as a separate nonlinear predictor for each class. It is designed to represent temporal structures of speech patterns as recognition cues. In particular, temporal correlation in successive feature vectors of a speech pattern is represented in the mappings formed as MLP input-output relations. Temporal distortion of speech is efficiently normalized by a dynamic-programming technique. Recognition and training algorithms are presented based on the combination of dynamic-programming and back-propagation techniques. Evaluation experiments were conducted using ten-digit vocabulary samples uttered by 107 speakers. A 99.8% recognition accuracy was obtained. This suggests that the model is effective for speaker-independent speech recognition. >

Complementary algorithms for the recognition of totally unconstrained handwritten numerals

Abstract: Two novel methods for recognizing totally unconstrained handwritten numerals are presented. One classifies samples based on structural features extracted from their skeletons; the other makes use of their contours. Both methods achieve high recognition rates (86.05%, 93.90%) and low substitution rates (2.25%, 1.60%). To take advantage of the inherent complementarity of the two methods, different ways of combining them are studied. It is shown that it is possible to reduce the substitution rate to 0.70%, while the recognition rate remains as high as 92.00% . Furthermore, if reliability is of utmost importance, one can avoid substitutions completely (reliability 100%) and still retain a fairly high recognition rate (84.85%). >

Small sample size effects in statistical pattern recognition: recommendations for practitioners and open problems

Abstract: The authors discuss the effects of sample size on the feature selection and error estimation for several types of classifiers. In addition to surveying prior work in this area, they give practical advice to today's designers and users of statistical pattern recognition systems. It is pointed out that one needs a large number of training samples if a complex classification rule with many features is being utilized. In many pattern recognition problems, the number of potential features is very large and not much is known about the characteristics of the pattern classes under consideration: thus, it is difficult to determine a priori the complexity of the classification rule needed. Therefore, even when the designer believes that a large number of training samples has been selected, they may not be enough for designing and evaluating the classification problem at hand. It is further noted that a small sample size can cause many problems in the design of a pattern recognition system. >

Two case studies in cost-sensitive concept acquisition

Abstract: This paper explores the problem of learning from examples when feature measurement costs are significant. It then extends two effective and familiar learning methods, ID3 and IBL, to address this problem. The extensions, CS-ID3 and CS-IBL, are described in detail and are tested in a natural robot domain and a synthetic domain. Empirical studies support the hypothesis that the extended methods are indeed sensitive to feature costs: they deal effectively with varying cost distributions and with irrelevant features.

Speaker-independent recognition of spoken English letters

Abstract: A description is presented of EAR, an English alphabet recognizer that performs speaker-independent recognition of letters spoken in isolation. During recognition, (a) signal processing routines transform the digitized speech into useful representations, (b) rules are applied to the representations to locate segment boundaries, (c) feature measurements are computed on the speech segments, and (d) a neural network uses the feature measurements to classify the letter. The system was trained on one token of each letter from 120 speakers. Performance was 95% when tested on a new set of 30 speakers. Performance was 96% when tested on a second token of each letter from the original 120 speakers. The recognition accuracy is 6% higher than that of previously reported systems. The high level of performance is attributed to accurate and explicit phonetic segmentation, the use of speech knowledge to select features that measure the important linguistic information, and the ability of the neural classifier to model the variability of the data

Self-organization of associative memory and pattern classification: recurrent signal processing on topological feature maps

Abstract: We extend the neural concepts of topological feature maps towards self-organization of auto-associative memory and hierarchical pattern classification. As is well-known, topological maps for statistical data sets store information on the associated probability densities. To extract that information we introduce a recurrent dynamics of signal processing. We show that the dynamics converts a topological map into an auto-associative memory for real-valued feature vectors which is capable to perform a cluster analysis. The neural network scheme thus developed represents a generalization of non-linear matrix-type associative memories. The results naturally lead to the concept of a feature atlas and an associated scheme of self-organized, hierarchical pattern classification.

Learning algorithms for neural networks with the Kalman filters

Abstract: Based on various approaches, several different learing algorithms have been given in the literature for neural networks. Almost all algorithms have constant learning rates or constant accelerative parameters, though they have been shown to be effective for some practical applications. The learning procedure of neural networks can be regarded as a problem of estimating (or identifying) constant parameters (i.e. connection weights of network) with a nonlinear or linear observation equation. Making use of the Kalman filtering, we derive a new back-propagation algorithm whose learning rate is computed by a time-varying Riccati difference equation. Perceptron-like and correlational learning algorithms are also obtained as special cases. Furthermore, a self-organising algorithm of feature maps is constructed within a similar framework.

Exploiting prior knowledge in network optimization: an illustration from medical prognosis

Abstract: Feed-forward layered networks trained on a pattern classification task in which the number of training patterns in each class is non-uniform, exhibit a strong classification bias towards those classes with largest membership. This is an unfortunate property of networks when the relative importance of classes with smaller membership is much greater than that of classes with many training patterns. In addition, there are many pattern classification tasks where different penalties are associated with misclassifying a pattern belonging to one class as another class. Generally, it is not known how to compensate for such effects in network training. This paper discusses an analytical regularization scheme whereby prior expectations of class importance occurring in the generalization data and misclassification costs may be incorporated into the training phase, thus compensating for the uneven and unfair class distributions occurring in the training set. The effects of the proposed scheme on the feature extractio...

Adding domain knowledge to SBL through feature construction

Abstract: This paper presents two methods for adding domain knowledge to similarity-based learning through feature construction, a form of representation change in which new features are constructed from relationships detected among existing features. In the first method, domain-knowledge constraints are used to eliminate less desirable new features before they are constructed. In the second method, domain-dependent transformations generalize new features in ways meaningful to the current problem. These two uses of domain knowledge are illustrated in CITRE where they are shown to improve hypothesis accuracy and conciseness on a tic-tat-toe classification problem.

Learning 3D object recognition strategies

Abstract: The problem of automatically learning knowledge-directed control strategies is considered. In particular, the authors address the problem of learning object-specific recognition strategies from object descriptions and sets of interpreted training images. A separate recognition strategy is developed for every object in the domain. The goal of each recognition strategy is to identify any and all instances of the object in an image, and give the 3-D position (relative to the camera) of each instance. The goal of the learning process is to build a strategy that minimizes the expected cost of recognition, subject to accuracy constraints imposed by the user. >

SIGART Member Dissertation Abstracts: Automatic Recognition of Machinable Features in Solid Models

Abstract: Recognition of machining features such as holes, slots and pockets is essential for the fully automatic manufacture of mechanical parts. This thesis discusses an experimental feature recognizer that uses a blend of artificial intelligence (AI) and computational geometry techniques. The recognizer is implemented in a rapid prototyping test bed consisting of the Knowledge Craft (TM) AI environment tightly coupled with the PADL-2 solid modeler, running under Unix on a SUN 3/260 computer. It is capable of finding features with interacting volumes (e.g., two crossing slots), and takes into account nominal shape information, tolerancing and other available data.

A texture classifier based on neural network principles

Abstract: A microprocessor-based system for texture classification and recognition is described. It is able to classify images containing stochastic textures. The maximum number of classes is currently 64. The learning and recognition are based on neural network principles. The topological feature map, a texture map, is created by self-organization. The recognition is based on learning vector quantization. A typical recognition rate for stochastic textures is 80% to 95%. The recognition rate depends on the number of classes and the quality of reference samples. New classes are easily taught by examples. The comparisons between stochastic textures is easy because of the texture map

Electronic musical instrument utilizing a neural network

Abstract: A musical tone parameter generating method and a musical tone generating device of this invention feature that when data inputted by a player is inputted into a neural network as input pattern, the neural network infers the parameters necessary to specify a musical tone wave form to be formed. This makes it possible to get parameters other than those stored in a memory by inferring, which increases variation of the musical tone to be generated.

A methodology for recognizing features in two-dimensional cylindrical part designs

Abstract: SUMMARY The interfacing of computer-aided design (CAD) to computer-aided manufacturing (CAM) is a vital step in automated manufacturing. An essential operation is the recognition of features from the part design. This paper presents a methodology for the recognition of features from two-dimensional rotational objects. First, this work defines the term ‘feature’ as a set of connected lines in the profile of the object, which satisfy certain geometric properties. Then, the task of feature recognition is decomposed into a set of distinct functions. These recognize, classify, decompose and reconstruct, and identify face sets which satisfy the definition of features. A prototype is developed which implements these functions. The important characteristics of this methodology are: (1) all cylindrical features are recognized, and most of them identified according to the input formats of a desired CAPP system; and (2) the system is modular and flexible and its functions can be easily modified.

Method and apparatus for time series signal recognition with signal variation proof learning

Abstract: A time series signal recognition capable of obtaining a high recognition rate even for the speech data with low S/N ratio in noisy environments. The time series signals are recognized by extracting a plurality of candidate feature vectors characterizing an individual time series signal, without fixing a boundary for the individual time series signal. Similarity values are calculated for each of the plurality of candidate feature vectors and the reference patterns stored in the recognition dictionary, from which one reference pattern for which the similarity value is greater than a prescribed threshold value is selected as a recognition result. New reference patterns to be stored in the recognition dictionary are learned by acquiring actual background noise of the apparatus, and mixing prescribed noiseless signal patterns with the acquired background noise to form signal patterns for learning. The signal patterns for learning are recognized by extracting features vectors for learning from the signal patterns for learning, and the new reference patterns are obtained from the extracted feature vectors for learning. The learning process is iterated at different noise levels, so as to optimize the determination of the word boundary. The background noise may be constantly acquired, and learning can be carried out using the noise data acquired immediately before the speech data is input.

Multi-phase recognition of multifont photoscript Arabic text

Abstract: A new system for the recognition of a multifont photoscript Arabic text is introduced. The distinguishing feature of such text is that it is written cursively. This imposes an additional requirement of isolating each character or set of overlapping characters before recognition. The proposed system is composed of three interleaved phases. The segmentation phase attempts to produce an initial set of characters from the connected text according to a set of predefined rules. The output is then passed to a preliminary classification phase that attempts to label the unknown characters into one of ten possible classes according to a set of rules that acquire their parameter values through learning. The last phase contains a more elaborate set of rules that recognize characters within each class. This recognition phase is designed to allow errors during segmentation and/or classification to be rectified through an adaptive recognition technique. The system has been implemented and tested on several fonts with a recognition rate of 130 words/min and an error rate of less than 6%. >

3D curved object recognition from multiple 2D camera views

Abstract: A new approach to 3D object recognition using multiple 2D camera views is proposed. The recognition system includes a turntable, a top camera, and a lateral camera. Objects are placed on the turntable for translation and rotation in the recognition process. 3D object recognition is accomplished by matching sequentially input 2D silhouette shape features against those of model shapes taken from a set of fixed camera views. This is made possible through the use of top-view shape centroids and principal axes for shape registration, as well as the use of a decision tree for feature comparison. The process is simple and efficient, involving no complicated 3D surface data computation and 3D object representation. The learning process can also be performed automatically. Good experimental results and fast recognition speed prove the feasibility of the proposed approach.

Self-organizing neural network character recognition on a massively parallel computer

Abstract: Two neural-network-based methods are combined to develop font-independent character recognition on a distributed array processor. Feature localization and noise reduction are achieved using least-squares optimized Gabor filtering. The filtered images are then presented to an ART-1-based learning algorithm which produces self-organizing sets of neural network weights used for character recognition. Implementation of these algorithms on a highly parallel computer with 1024 processors allows high-speed character recognition to be achieved in 8 ms/image with greater than 99% accuracy on machine print and 80% accuracy on unconstrained hand-printed characters

A comparative study of neural network algorithms applied to optical character recognition

Abstract: Three simple general purpose networks are tested for pattern classification on an optical character recognition problem. The feed-forward (multi-layer perceptron) network, the Hopfield network and a competitive learning network are compared. The input patterns are obtained by optically scanning images of printed digits and uppercase letters. The resulting data is used as input for the networks with two-state input nodes; for others, features are extracted by template matching and pixel counting. The classification capabilities of the networks are compared with a nearest neighbour algorithm applied to the same feature vectors. The feed-forward network reaches the same recognition rates as the nearest neighbour algorithm, even when only a small percentage of the possible connections is used. The Hopfield network performs less well, and overloading of the network remains a problem. Recognition rates with the competitive learning network, if input patterns are clustered well, are again as high as the nearest neighbour algorithm.