Showing papers on "Feature (machine learning) published in 1987"

System for self-organization of stable category recognition codes for analog input patterns

Abstract: A neural network includes a feature representation field which receives input patterns. Signals from the feature representation field select a category from a category representation field through a first adaptive filter. Based on the selected category, a template pattern is applied to the feature representation field, and a match between the template and the input is determined. If the angle between the template vector and a vector within the representation field is too great, the selected category is reset. Otherwise the category selection and template pattern are adapted to the input pattern as well as the previously stored template. A complex representation field includes signals normalized relative to signals across the field and feedback for pattern contrast enhancement.

Hybrid methods in pattern recognition

Abstract: The field of pattern recognition has grown enormously in recent years and a wide variety of techniques have been developed for various applications. Traditionally, these techniques can be categorized into statistical, or decision theoretic, and structural methods. Additionally, artificial intelligence based approaches have become very important recently. Each of the different methods has its strength and its limitations. For overcoming these limitations, statistical, structural, and artificial intelligence based methods are mixed sometimes. This results in a hybrid approach.

Cognitive dissonance and utility maximization: A general framework

Abstract: Neoclassical theory of utility maximization assumes irrational behavior to be unsystematic and therefore impossible to model. Recent advances in behavioral decision theory suggests irrationality may be systematic. In line with these and earlier findings from the theory of cognitive dissonance, a simple descriptive model of utility maximization is developed with the added feature of an information filter. The model is then used to explain a few ‘irrational’ micro and macro behaviors.

Comparison of rule-building expert systems with pattern recognition for the classification of analytical data

Abstract: Two expert systems of the rule-building type, TIMM and EX-TRAN, are compared with pattern recognition methods for the classification of olive oils of different origins. Both expert systems are more user-friendly than the pattern recognition programs and TIMM yields slightly better results than nearest neighbors classifiers and linear discriminant analysis.

Learning phonetic features using connectionist networks

Abstract: A method for learning phonetic features from speech data using connectionist networks is described. A temporal flow model is introduced in which sampled speech data flows through a parallel network from input to output units. The network uses hidden units with recurrent links to capture spectral/temporal characteristics of phonetic features. A supervised learning algorithm is presented which performs gradient descent in weight space using a coarse approximation of the desired output as an evaluation function. A simple connectionist network with recurrent links was trained on a single instance of the word pair "no" and "go", and successful learned a discriminatory mechanism. The trained network also correctly discriminated 98% of 25 other tokens of each word by the same speaker. A single integrated spectral feature was formed without segmentation of the input, and without a direct comparison of the two items.

Dynamic speaker adaptation for feature-based isolated word recognition

Abstract: In this paper, we describe efforts to improve the performance of FEATURE, the Carnegie-Mellon University speaker-independent speech recognition system that classifies isolated letters of the English alphabet by enabling the system to learn the acoustical characteristics of individual speakers. Even when features are designed to be speaker-independent, it is frequently observed that feature values may vary more from speaker to speaker for a single letter than they vary from letter to letter. In these cases, it is necessary to adjust the system's statistical description of the features of individual speakers to obtain improved recognition performance. This paper describes a set of dynamic adaptation procedures for updating expected feature values during recognition. The algorithm uses maximum a posteriori probability (MAP) estimation techniques to update the mean vectors of sets of feature values on a speaker-by-speaker basis. The MAP estimation algorithm makes use of both knowledge of the observations input to the system from an individual speaker and the relative variability of the features' means within and across all speakers. In addition, knowledge of the covariance of the features' mean vectors across the various letters enables the system to adapt its representation of similar-sounding letters after any one of them is presented to the classifier. The use of dynamic speaker adaptation improves classification performance of FEATURE by 49 percent after four presentations of the alphabet, when the system is provided with supervised training indicating which specific utterance had been presented to the classifier from a particular user. Performance can be improved by as much as 31 percent when the system is allowed to adapt passively in an unsupervised learning mode. without any information from individual users.

Color representation using statistical pattern recognition.

Abstract: A statistical pattern recognition method, the subspace method, closely related to the principal component analysis was used in color recognition. It is shown that color spectra can be accurately reconstructed using a few principal spectra. Further, it is shown that our method is capable of discriminating samples which were inseparable using usual methods.

Recognition of characteristic patterns in sets of functionally equivalent DNA sequences

Abstract: An algorithm has been developed for the identification of unknown patterns which are distinctive for a set of short DNA sequences believed to be functionally equivalent. A pattern is defined as being a string, containing fully or partially specified nucleotides at each position of the string. The advantage of this 'vague' definition of the pattern is that it imposes minimum constraints on the characterization of patterns. A new feature of the approach developed here is that it allows a 'fair' simultaneous testing of patterns of all degrees of degeneracy. This analysis is based on an evaluation of inhomogeneity in the empirical occurrence distribution of any such pattern within a set of sequences. The use of the nonparametric kernel density estimation of Parzen allows one to assess small disturbances among the sequence alignments. The method also makes it possible to identify sequence subsets with different characteristic patterns. This algorithm was implemented in the analysis of patterns characteristic of sets of promoters, terminators and splice junction sequences. The results are compared with those obtained by other methods.

Speech recognition by acoustic/phonetic system and technique

Abstract: A speech recognition system and technique of the acoustic/phonetic type is made speaker-independent and capable of continuous speech recognition during fluent discourse by a combination of techniques which include, inter alia, using a so-called continuously-variable-duration hidden Markov vodel in identifying word segments, i.e., phonetic units, and developing proposed phonetic sequences by a durationally-responsive recursion before any lexical access is attempted. Lexical access is facilitated by the phonetic transcriptions provided by the durationally-responsive recursion; and the resulting array of word candidates facilitates the subsequent alignment of the word candidates with the acoustic feature signals. A separate step is used for aligning the members of the candidate word arrays with the acoustic feature signals representative of the corresponding portion of the utterance. Any residual work selection ambiguities are then more readily resolved, regardless of the ultimate sentence selection technique employed.

Feature analysis for symbol recognition by elastic matching

Abstract: A technique has been developed for the recognition of unconstrained handwritten discrete symbols based on elastic matching against a set of prototypes generated by individual writers. The incorporation of feature analysis with elastic matching to eliminate unlikely prototypes is presented in this paper and is shown to greatly reduce the required processing time without any deterioration in recognition performance.

Speech recognition system using vector quantization

Abstract: PURPOSE: To reduce learning samples and to shorten calculation time by using separative vector quantization for individually generating a code book in each feature as vector quantization and executing individual vector quantization. CONSTITUTION: A plurality of code books corresponding to the sorts of features of input voices are prepared, respective code book are quantized and recognition is executed by using a plurality of found code strings. Namely a voice signal is amplified by an amplifier, a return noise is removed by a low pass filter 2, the noise-removed voice signal is converted into a digital signal by an A/D converter 3 and the feature of the voice is extracted by a computer 5. A feature string in each extracted feature is collated with an already stored reference pattern by a matching part, based upon a splite method, the matching distance is sent to a result judging part 5, whether the result is suitable for a recognition candidate or not is judged, and the recognition result is outputted. Consequently, learning samples can be reduced and calculation volume can be reduced. COPYRIGHT: (C)1989,JPO

Learning and using specific instances.

Abstract: A biologically plausible method for rapidly learning specific instances is described. It is contrasted with a formal model of classical conditioning (Rescorla-Wagner learning/perception training), which is shown to be relatively good for learning generalizations, but correspondingly poor for learning specific instances. A number of behaviorally relevant applications of specific instance learning are considered. For category learning, various combinations of specific instance learning and generalization are described and analyzed. Two general approaches are considered: the simple inclusion of Specific Instance Detectors (SIDs) as additional features during perception training, and specialized treatment in which SID-based categorization takes precedence over generalization-based categorization. Using the first approach, analysis and empirical results demonstrate a potential problem in representing feature presence and absence in a symmetric fashion when the frequencies of feature presence and absence are very different. However, it is shown that by using the proper representation, the addition of SIDs can only improve the convergence rate of perceptron training, the greatest improvement being achieved when SIDs are preferentially allocated for peripheral positive and negative instances. Some further improvement is possible if SIDs are treated in a specialized manner.

A syntactic omni-font character recognition system

Abstract: This paper introduces a syntactic omni-font character recognition system. The "omni-font" attribute reflects the wide range of fonts that fall within the class of characters that can be recognized. This includes hand-printed characters as well. A structural pattern-matching approach is employed. Essentially, a set of loosely constrained rules specify pattern components and their interrelationships. The robustness of the system is derived from the orthogonal set of pattern descriptors, location functions, and the manner in which they are combined to exploit the topological structure of characters. By virtue of the new pattern description language, PDL, developed in this paper, the user may easily write rules to define new patterns for the system to recognize. The system also features scale-invariance and user-definable sensitivity to tilt orientation.

The Compositional Nature of (In)Definiteness

Abstract: In certain syntactic contexts, only a restricted set of determiners can be inserted. This restriction is often associated with the feature of (in)definiteness. However, the extensive literature on (in)definiteness appears to lack a clear-cut definition of the categories to which this feature applies. A comparison of the distributional restrictions in two contexts that are assumed to be sensitive to (in)definiteness shows this. The contexts are (a) sentences with initial 'there' which have been extensively studied by Milsark (1977) and (b) partitive NP-constructions (as analysed by e.g. Barwise & Cooper (1981)). The classification of determimers to be proposed in this article will be based on a system incorporating *two* features instead of one. Empirical evidence for this proposal will be derived from the analysis of several other restrictive contexts.

Information recognizing device

Abstract: PURPOSE:To recognize characters written in the running hand even if the number of character patterns stored in a character dictionary table is small, by absorbing variation of the number of strokes and a pattern even if it is generated in a partial pattern, by an identification operation in a partial identification means CONSTITUTION:As for an input character pattern from an information input part 1, the feature of the number of strokes and a feature point coordinate of each stroke, etc is extracted by a feature extracting means 3, and when this feature is inputted, a partial identification means 5 compares the input character pattern and all partial patterns which have been registered in advance A character identification means 8 limits the number of strokes of all character patterns with regard to the partial pattern which has been identified by the identification means 5, and decides whether they are similar or not In this way, prior to comparison of an input stroke string and a pattern which has been registered in a character dictionary table 7, classification is executed by deriving the similarly of the input stroke string and the partial pattern of the character which has been determined in advance, the processing quantity is decreased remarkably, also, especially variation of the number of strokes by characters written in the running hand in a sub-pattern are corrected and many characters can be recognized without increasing the capacity of a dictionary table

Optical analog of two-dimensional neural networks and their application in recognition of radar targets

Abstract: Optical analogs of 2−D distribution of idealized neurons (2−D neural net) based on partitioning of the resulting 4−D connectivity matrix are discussed. These are desirable because of compatibility with 2−D feature spaces and ability to realize denser networks. An example of their use with sinogram classifiers derived from ralistic radar data of scale models of three aerospace objects as learning set is given. Super‐resolved recognition from partial information that can be as low as 20% of the sinogram data is demonstrated together with a capacity for error correction and generalization.

Speech recognition system.

Abstract: The present invention relates to a speech recognition system of the type which comprises storage means (10, 11) for storing selected parameters for each of a plurality of words in a vocabulary to be used for recognition of an input item of speech, comparison means (42) for comparing parameters of each unknown word in an input item of speech with the stored parameters, and indication means (12, 46) responsive to the result of the comparison operation for indicating which of the plurality of vocabulary words most closely resembles each unknown input word According to the invention the speech recognition system is characterised in that the stored parameters comprise for each vocabulary word a set of labels each representing a feature of the vocabulary word occurring at a respective segmentation point in the vocabulary word and the probability of the feature associated with each label occurring at a segmentation point in a word Further, the comparison means compares the stored sets of parameters with a set of labels for each unknown input word each representing a feature of the unknown input word occurring at a respective segmentation point in the unknown input word

System for continuous speech recognition through transition networks

Abstract: Phoneme feature parameters are extracted from input digital speech signals by means of LPC analysis. Phonetic segments having phonetical meanings are obtained together with similarities to prescribed basic phonetic segments from the feature parameters to be passed through nodes of transition networks provided for each word. In passing the nodes, scores for similarity Sj of predetermined segments of the corresponding phonetic segments are made in selective scoring and the accumulation of the scores is used for recognition of continuous word speech.

Self-organizing neural network models for visual pattern recognition.

Abstract: Two neural network models for visual pattern recognition are discussed. The first model, called a “neocognitron”, is a hierarchical multilayered network which has only afferent synaptic connections. It can acquire the ability to recognize patterns by “learning-without-a-teacher”: the repeated presentation of a set of training patterns is sufficient, and no information about the categories of the patterns is necessary. The cells of the highest stage eventually become “gnostic cells”, whose response shows the final result of the pattern-recognition of the network. Pattern recognition is performed on the basis of similarity in shape between patterns, and is not affected by deformation, nor by changes in size, nor by shifts in the position of the stimulus pattern.

Speech recognition system which avoids ambiguity when matching frequency spectra by employing an additional verbal feature

Abstract: In a speech recognition system using normalized spectrum matching, a second feature pattern is calculated and compared with reference patterns. The similarity obtained as a result of the comparison is used to determine overall similarity, from which the recognition is made. The second feature pattern can be a spectrum variation pattern, a level decrease pattern, or a spectrum relative value pattern.

Word voice recognition equipment

Abstract: PURPOSE: To improve the recognition rate of word recognition by preparing the syllable pattern of a double consonant 'tsu' from a consonant section or a buzz sound section of a single syllable pattern and selecting and using the syllable pattern in accordance with a syllable pattern to be connected. CONSTITUTION: An operator's voice is patterned by an input part 1, separated into different syllables and inputted to a syllable templete dictionary part 2. A mean value calculation part 8 calculates the mean value of the features of consonants or the mean value of all feature values in a prescribed section having a remarkable buzz sound in buzz sound and consonant extracting sections and stores the mean value in a double consonant pattern dictionary 9. Then the patterns stored in the dictionary parts 2, 9 and a reference pattern in a word dictionary part 3 are inputted to a synthetic pattern preparing part 4. The preparing part 4 prepares a synthetic pattern from the input patterns in accordance with a rule stored in a double consonant pattern selection rule 10 and stores the prepared synthetic pattern in a synthetic pattern dictionary part 5. The synthetic pattern improves the probability of collation with an input voice in a pattern collating part 6 at the time of word recognition.

Absolutely stable learning of recognition codes by a self-organizing neural network

Abstract: A neural network which self‐organizes and self‐stabilizes its recognition codes in response to arbitrary orderings of arbitrarily many and arbitrarily complex binary input patterns is here outlined. Top‐down attentional and matching mechanisms are critical in self‐stabilizing the code learning process. The architecture embodies a parallel search scheme which updates itself adaptively as the learning process unfolds. After learning self‐stabilizes, the search process is automatically disengaged. Thereafter input patterns directly access their recognition codes, or categories, without any search. Thus recognition time does not grow as a function of code complexity. A novel input pattern can directly access a category if its shares invariant properties with the set of familiar exemplars of that category. These invariant properties emerge in the form of learned critical feature patterns, or prototypes. The architecture possesses a context‐sensitive self‐scaling property which enables its emergent critical fea...

The role of structural context in perception: Syntax in the recognition of algebraic expressions

Abstract: Two character-identification experiments investigated the function of structural context during the processing of briefly exposed algebraic strings. Neither experiment prodded evidence to support the notion of an algebra-superiority effect, a contextually driven enhancement of the recognition of specific algebraic characters. However, the results of Experiment 2 indicate that the structure of algebra does provide information at the level of a character’s categorical denomination. These findings suggest that the parsing of an algebraic string includes a level of processing in which its structural context places restrictions on the denominations of its symbols. A processing model of algebraic perception is proposed that incorporates these syntactic constraints—constraints that appear to be independent of feature-based character identification processes.

Syntactic pattern recognition and Hough transformation for reconstruction of seismic patterns

Abstract: Hierarchical syntactic pattern recognition and the Hough transformation are proposed for automatic recognition and reconstruction of seismic patterns in seismograms. In the first step, the patterns are hierarchically decomposed or recognized into single patterns, straight‐line patterns, or hyperbolic patterns, using syntactic pattern recognition. In the second step, the Hough transformation technique is used for reconstruction, pattern by pattern. The system of syntactic seismic pattern recognition includes envelope generation, a linking process in the seismogram, segmentation, primitive recognition, grammatical inference, and syntax analysis. The seismic patterns are automatically recognized and reconstructed.

Pattern feature extracting system

Abstract: In a speech signal processing system for transmission or recognition, cepstrum data is normalized by subtraction from its straight-line approximation. As a result, personal and transmission characteristics are eliminated, and both data and subtraction computations are reduced.

Learning phonetic features using connectionist networks

Abstract: A method for learning phonetic features from speech data using connectionist networks is described. A temporal flow model is introduced in which sampled speech data flow through a parallel network from input to output units. The network uses hidden units with recurrent links to capture spectral/temporal characteristics of phonetic features. A supervised learning algorithm is presented which performs gradient descent in weight space using a course approximation of the desired output as an evaluation function. A simple connectionist network with recurrent links was trained on a single instance of the work pair “no” and “go,” and successfully learned a discriminatory mechanism. The trained network also correctly discriminated 98% of 25 other tokens of each word by the same speaker. The discriminatory feature was formed without segmentation of the input, and without a direct comparison of the two items. The network formed an internal representation of a single, integrated spectral feature which has a theoretical basis in human acoustic‐phonetic perception.