Neural networks for improved target differentiation and localization with sonar.

This study investigates the processing of sonar signals using neural networks for robust differentiation of commonly encountered features in indoor robot environments. Differentiation of such features is of interest for intelligent systems in a variety of applications. Different representations of amplitude and time-of-flight measurement patterns acquired from a real sonar system are processed. In most cases, best results are obtained with the low-frequency component of the discrete wavelet transform of these patterns. Modular and non-modular neural network structures trained with the back-propagation and generating-shrinking algorithms are used to incorporate learning in the identification of parameter relations for target primitives. Networks trained with the generating-shrinking algorithm demonstrate better generalization and interpolation capability and faster convergence rate. Neural networks can differentiate more targets employing only a single sensor node, with a higher correct differentiation percentage (99%) than achieved with previously reported methods (61-90%) employing multiple sensor nodes. A sensor node is a pair of transducers with fixed separation, that can rotate and scan the target to collect data. Had the number of sensing nodes been reduced in the other methods, their performance would have been even worse. The success of the neural network approach shows that the sonar signals do contain sufficient information to differentiate all target types, but the previously reported methods are unable to resolve this identifying information. This work can find application in areas where recognition of patterns hidden in sonar signals is required. Some examples are system control based on acoustic signal detection and identification, map building, navigation, obstacle avoidance, and target-tracking applications for mobile robots and other intelligent systems.     

A Position Identification and Path Labelling Mechanism for a Neural Model of Visual Awareness

We discuss a mechanism for the representation of location and path references by purely neural means in a digital model of visual awareness. The underlying base awareness model features a “virtual eye” that can look out into a virtual world populated with objects to be recognised. The presented models take in to account location information sensed this virtual eye, as it roves across portions of the sensed world, to identify and represent position information regarding the sensed objects. We further discuss the addition of neural layer structures to the base awareness model to process output from the virtual eye and accordingly handle location and path identification and labelling is described. Planned applications and the context of the modelling of location and path recognition in the context of motion verb recognition are then considered. It is accordingly seen that this work enables a visual awareness model to identify and represent by purely neural means a trajectory for motion in its sensed world.     

Low-variance Multitaper Mel-frequency Cepstral Coefficient Features for Speech and Speaker Recognition Systems

In this paper, we investigate low-variance multitaper spectrum estimation methods to compute the mel-frequency cepstral coefficient (MFCC) features for robust speech and speaker recognition systems. In speech and speaker recognition, MFCC features are usually computed from a single-tapered (e.g., Hamming window) direct spectrum estimate, that is, the squared magnitude of the Fourier transform of the observed signal. Compared with the periodogram, a power spectrum estimate that uses a smooth window function, such as Hamming window, can reduce spectral leakage. Windowing may help to reduce spectral bias, but variance often remains high. A multitaper spectrum estimation method that uses well-selected tapers can gain from the bias-variance trade-off, giving an estimate that has small bias compared with a single-taper spectrum estimate but substantially lower variance. Speech recognition and speaker verification experimental results on the AURORA-2 and AURORA-4 corpora and the NIST 2010 speaker recognition evaluation corpus (telephone as well as microphone speech), respectively, show that the multitaper methods perform better compared with the Hamming-windowed spectrum estimation method. In a speaker verification task, compared with the Hamming window technique, the sinusoidal weighted cepstrum estimator, multi-peak, and Thomson multitaper techniques provide a relative improvement of 20.25, 18.73, and 12.83 %, respectively, in equal error rate.     

Bifurcation Properties for Fractional Order Delayed BAM Neural Networks

In the past several decades, many papers involving the stability and Hopf bifurcation of delayed neural networks have been published. However, the results on the stability and Hopf bifurcation for fractional order neural networks with delays and fractional order neural networks with leakage delays are very rare. This paper is concerned with the stability and the existence of Hopf bifurcation of fractional order BAM neural networks with or without leakage delay. The Laplace transform, stability and bifurcation theory of fractional-order differential equations and Matlab software will be applied. The stability condition and the sufficient criterion of existence of Hopf bifurcation for fractional order BAM neural networks with delay (leakage delay) are established. It is found that when the sum of two delays (leakage delay) crosses a critical value, then a Hopf bifurcation will appear. The obtained results play an important role in designing neural networks. Also the derived results are new and enrich the bifurcation theory of fractional order delayed differential equations.

     

Percevoir un visage dans la petite enfance

Faces are very complex objects. Face processing is a set of sophisticated skills that serve several mental functions (species recognition, recognition of others as similar to self and vice-versa; idendity, emotional expressions, communication signals recognition, affective position, attachment, and so on). A part of neural networks and mechanisms that are involved in face processing belong to different levels of neural integration and are also involved in other object processing (including processing of body and body parts). Another part is involved in social and affective competences and attachment. It is assumed that among neural computations involved in face processing, some are face specific (and might be common to all primates), others are less specific, some are grossly primed from birth and later, and their development results from progressive specialisation through interaction with the perceptual and social environment. Uncovering the face ‘primers’ and disentangling the complex set of interactions between brain and environment is the task of developmental studies.     

A category-specific deficit of spatial representation: the case of autotopagnosia

Following a vascular lesion in the parietal cortex of the language dominant hemisphere (right in one case), two patients showed a striking dissociation between spared naming, recognition and use of their body parts and an inability in localising on verbal command the same body parts on themselves and on a mannequin (Autotopagnosia, AT). The patients were submitted to a modified version of Reed and Farah Test (1995), a test that taps the ability to encode changes of body position as opposed to changes of position of objects. Their performance differed from normal controls, showing a specific deficit in encoding body position.It is suggested that AT could be the consequence of a lesion in a specific neural circuit, located in the language dominant hemisphere, whose function is to encode the body position for both oneself and others.     

Modulation of theta phase synchronization in the human electroencephalogram during a recognition memory task

To the extent that recognition memory relies on interactions among widely distributed neural assemblies across the brain, phase synchronization between brain rhythms may play an important role in meditating those interactions. As the theta rhythm is known to modulate in power during the recognition memory process, we aimed to determine how the phase synchronization of the theta rhythms across the brain changes with recognition memory. Fourteen human participants performed a visual object recognition task in a virtual reality environment. Electroencephalograms were recorded from the scalp of the participants while they either recognized objects that had been presented previously or identified new objects. From the electroencephalogram recordings, we analyzed the phase-locking value of the theta rhythms, which indicates the degree of phase synchronization. We found that the overall phase-locking value recorded during the recognition of previously viewed objects was greater than that recorded during the identification of new objects. Specifically, the theta rhythms became strongly synchronized between the frontal and the left parietal areas during the recognition of previously viewed objects. These results suggest that the recognition memory process may involve an interaction between the frontal and the left parietal cortical regions mediated by theta phase synchronization.     

The sonar aperture and its neural representation in bats

As opposed to visual imaging, biosonar imaging of spatial object properties represents a challenge for the auditory system because its sensory epithelium is not arranged along space axes. For echolocating bats, object width is encoded by the amplitude of its echo (echo intensity) but also by the naturally covarying spread of angles of incidence from which the echoes impinge on the bat's ears (sonar aperture). It is unclear whether bats use the echo intensity and/or the sonar aperture to estimate an object's width. We addressed this question in a combined psychophysical and electrophysiological approach. In three virtual-object playback experiments, bats of the species Phyllostomus discolor had to discriminate simple reflections of their own echolocation calls differing in echo intensity, sonar aperture, or both. Discrimination performance for objects with physically correct covariation of sonar aperture and echo intensity ("object width") did not differ from discrimination performances when only the sonar aperture was varied. Thus, the bats were able to detect changes in object width in the absence of intensity cues. The psychophysical results are reflected in the responses of a population of units in the auditory midbrain and cortex that responded strongest to echoes from objects with a specific sonar aperture, regardless of variations in echo intensity. Neurometric functions obtained from cortical units encoding the sonar aperture are sufficient to explain the behavioral performance of the bats. These current data show that the sonar aperture is a behaviorally relevant and reliably encoded cue for object size in bat sonar.     

Recognition of objects in non-canonical views: a functional MRI study.

BACKGROUND: The neural correlate of object recognition in non-canonical views is uncertain, but there is evidence for involvement of neural pathways, possibly separate from those used for object recognition in canonical views. METHODS: Boxcar functional MRI (fMRI) techniques were used to detect neural activity while eight normal subjects were instructed to identify digital photographs of objects in non-canonical and canonical orientations. RESULTS: The right angular gyrus, the left inferior temporal gyrus, and the right cerebellum showed significant fMRI activity during non-canonical as opposed to canonical viewing. CONCLUSIONS: Subjects recognizing objects in non-canonical orientations engage in a process separate from, or in addition to, the process used in recognizing objects in canonical orientations.     

Induced gamma band responses predict recognition delays during object identification

Neural mechanisms of object recognition seem to rely on activity of distributed neural assemblies coordinated by synchronous firing in the gamma-band range (>20 Hz). In the present electroencephalogram (EEG) study, we investigated induced gamma band activity during the naming of line drawings of upright objects and objects rotated in the image plane. Such plane-rotation paradigms elicit view-dependent processing, leading to delays in recognition of disoriented objects. Our behavioral results showed reaction time delays for rotated, as opposed to upright, images. These delays were accompanied by delays in the peak latency of induced gamma band responses (GBRs), in the absence of any effects on other measures of EEG activity. The latency of the induced GBRs has thus, for the first time, been selectively modulated by an experimental manipulation that delayed recognition. This finding indicates that induced GBRs have a genuine role as neural markers of late representational processes during object recognition. In concordance with the view that object recognition is achieved through dynamic learning processes, we propose that induced gamma band activity could be one of the possible cortical markers of such dynamic object coding.     

A Preferential Role for Ventromedial Prefrontal Cortex in Assessing “the Value of the Whole” in Multiattribute Object Evaluation

Everyday decision-making commonly involves assigning values to complex objects with multiple value-relevant attributes. Drawing on object recognition theories, we hypothesized two routes to multiattribute evaluation: assessing the value of the whole object based on holistic attribute configuration or summing individual attribute values. In two samples of healthy human male and female participants undergoing eye tracking and functional magnetic resonance imaging (fMRI) while evaluating novel pseudo objects, we found evidence for both forms of evaluation. Fixations to and transitions between attributes differed systematically when the value of pseudo objects was associated with individual attributes or attribute configurations. Ventromedial prefrontal cortex (vmPFC) and perirhinal cortex were engaged when configural processing was required. These results converge with our recent findings that individuals with vmPFC lesions were impaired in decisions requiring configural evaluation but not when evaluating the sum of the parts. This suggests that multiattribute decision-making engages distinct evaluation mechanisms relying on partially dissociable neural substrates, depending on the relationship between attributes and value.SIGNIFICANCE STATEMENT Decision neuroscience has only recently begun to address how multiple choice-relevant attributes are brought together during evaluation and choice among complex options. Object recognition research makes a crucial distinction between individual attribute and holistic/configural object processing, but how the brain evaluates attributes and whole objects remains unclear. Using fMRI and eye tracking, we found that the vmPFC and the perirhinal cortex contribute to value estimation specifically when value was related to whole objects, that is, predicted by the unique configuration of attributes and not when value was predicted by the sum of individual attribute values. This perspective on the interactions between subjective value and object processing mechanisms provides a novel bridge between the study of object recognition and reward-guided decision-making.     

Fully complex magnetoencephalography

Complex numbers appear naturally in biology whenever a system can be analyzed in the frequency domain, such as physiological data from magnetoencephalography (MEG). For example, the MEG steady state response to a modulated auditory stimulus generates a complex magnetic field for each MEG channel, equal to the Fourier transform at the stimulus modulation frequency. The complex nature of these data sets, often not taken advantage of, is fully exploited here with new methods. Whole-head, complex magnetic data can be used to estimate complex neural current sources, and standard methods of source estimation naturally generalize for complex sources. We show that a general complex neural vector source is described by its location, magnitude, and direction, but also by a phase and by an additional perpendicular component. We give natural interpretations of all the parameters for the complex equivalent-current dipole by linking them to the underlying neurophysiology. We demonstrate complex magnetic fields, and their equivalent fully complex current sources, with both simulations and experimental data.     

Perceptual and Semantic Representations at Encoding Contribute to True and False Recognition of Objects

When encoding new episodic memories, visual and semantic processing is proposed to make distinct contributions to accurate memory and memory distortions. Here, we used fMRI and preregistered representational similarity analysis to uncover the representations that predict true and false recognition of unfamiliar objects. Two semantic models captured coarse-grained taxonomic categories and specific object features, respectively, while two perceptual models embodied low-level visual properties. Twenty-eight female and male participants encoded images of objects during fMRI scanning, and later had to discriminate studied objects from similar lures and novel objects in a recognition memory test. Both perceptual and semantic models predicted true memory. When studied objects were later identified correctly, neural patterns corresponded to low-level visual representations of these object images in the early visual cortex, lingual, and fusiform gyri. In a similar fashion, alignment of neural patterns with fine-grained semantic feature representations in the fusiform gyrus also predicted true recognition. However, emphasis on coarser taxonomic representations predicted forgetting more anteriorly in the anterior ventral temporal cortex, left inferior frontal gyrus and, in an exploratory analysis, left perirhinal cortex. In contrast, false recognition of similar lure objects was associated with weaker visual analysis posteriorly in early visual and left occipitotemporal cortex. The results implicate multiple perceptual and semantic representations in successful memory encoding and suggest that fine-grained semantic as well as visual analysis contributes to accurate later recognition, while processing visual image detail is critical for avoiding false recognition errors.SIGNIFICANCE STATEMENT People are able to store detailed memories of many similar objects. We offer new insights into the encoding of these specific memories by combining fMRI with explicit models of how image properties and object knowledge are represented in the brain. When people processed fine-grained visual properties in occipital and posterior temporal cortex, they were more likely to recognize the objects later and less likely to falsely recognize similar objects. In contrast, while object-specific feature representations in fusiform gyrus predicted accurate memory, coarse-grained categorical representations in frontal and temporal regions predicted forgetting. The data provide the first direct tests of theoretical assumptions about encoding true and false memories, suggesting that semantic representations contribute to specific memories as well as errors.     

Dimension change, coarse grained coding and pattern recognition in spatio-temporal nonlinear systems

Several research programs employing spatio-temporal recurrent dynamics and changes in dimensionality have extended the dialog on neural computation and coding beyond classical frameworks such as feed forward and attractor neural networks and feature detectors. Some have emphasized spiking networks, while others emphasize oscillations and synchronization as the locus of computation and coding. In this paper, the formalism of locally connected homogeneous coupled map lattices is described. Its deployment in an extended version of the dynamical recognizer framework is described, and is compared with density coding, computational mechanics, and liquid state machine frameworks for neural computation. A population coding strategy based on coarse graining the continuous valued distribution of all sites in the lattice is developed and examined as a form of dimension reduction. Results on recognition of 3-D objects are reported. In order to better understand the dynamics supporting recognition, measures suggested by these other research programs and computational frameworks were examined. Dynamics trajectories from object recognition trials were examined for correlation with recognition rates and measures of the distance of the representation space statistics between the target objects and noise initial conditions, and the intrinsic separation between different objects in the set to be classified were performed. These results raise questions about the efficacy of density coding as an explanation for the results, and on the validity of recent criticisms that chaotic systems cannot satisfy separation requirements required for real time computation.     

A Hierarchical Predictive Coding Model of Object Recognition in Natural Images

Predictive coding has been proposed as a model of the hierarchical perceptual inference process performed in the cortex. However, results demonstrating that predictive coding is capable of performing the complex inference required to recognise objects in natural images have not previously been presented. This article proposes a hierarchical neural network based on predictive coding for performing visual object recognition. This network is applied to the tasks of categorising hand-written digits, identifying faces, and locating cars in images of street scenes. It is shown that image recognition can be performed with tolerance to position, illumination, size, partial occlusion, and within-category variation. The current results, therefore, provide the first practical demonstration that predictive coding (at least the particular implementation of predictive coding used here; the PC/BC-DIM algorithm) is capable of performing accurate visual object recognition.     

Spatial and temporal proximity as factors in shape recognition

Prior research from this laboratory examined minimal stimulus conditions that allow for recognition of objects. Using briefly flashed dots that marked the outer border of objects, it was found that timing differentials within and among successive dot pairs affected recognition, with significant declines being seen by the addition of temporal separations in the millisecond range. These experiments were done with dot pairs that had close spatial proximity, which leaves open the possibility that the effects could be attributed to strictly local neural encoding processes. The present research reports that spatial separation of pair members resulted in declines in recognition that were similar to those produced with close spacing of pair members. Both for close and separated dot pairs, recognition was best when they were displayed with near simultaneity, which likely generated synchronized spikes in the retina. These results provide cognitive evidence in support of proposals that synchronous neural activity is part of the image encoding process. The physiological literature is surveyed and discussed in an effort to delineate the issues, and a tentative model of retinal response to these stimulus conditions is offered.     

Fat neural network for recognition of position-normalised objects

The design of a recognition system for natural objects is difficult, mainly because such objects are subject to a strong variability that cannot be easily modelled: planktonic species possess such highly variable forms. Existing plankton recognition systems usually comprise feature extraction processing upstream of a classifier. Drawbacks of such an approach are that the design of relevant feature extraction processes may be very difficult, especially if classes are numerous and if intra-class variability is high, so that the system becomes specific to the problem for which features have been tuned. The opposite course that we take is based on a structured multi-layer neural network with no shared weights, which generates its own features during training. Such a large parameterised—fat—network exhibits good generalisation capabilities for pattern recognition problems dealing with position-normalised objects, even with as many as one thousand weights as training examples. The advantage of such large networks, in terms of generalisation efficiency, adaptability and classification time, is demonstrated by applying the network to three plankton recognition and face recognition problems. Its ability to perform good generalisation with few training examples, but many weights, is an open theoretical problem.     

Apperceptive agnosia and face recognition

Abstract

Are faces and objects recognized by separate visual recognition systems or might a single system subserve the recognition of both classes of input? Recognition of faces and objects by a single system predicts that prosopagnosics, who selectively lose the ability to recognize faces due to brain damage, should also lose the ability to recognize objects. Contrary to this prediction, case studies of prosopagnosia have reported intact object recognition. Further support for separate visual recognition systems comes from the case of HH reported here. Following a stroke involving the left posterior cortex, HH has a severe apperceptive visual agnosia for visually presented objects and an alexia for words. Yet, he shows relatively spared visual face processing. Such a performance pattern completes a double dissociation between face and object processing when coupled with prosopaganosia. More importantly, HH is the first apperceptive visual object agnosic to demonstrate spared face processing. The severity of his object-processing deficit is such that from the earliest levels in the visual processing hierarchy, distinct neural substrates must be responsible for processing some objects and faces. These results are discussed as support for Farah's model (Visual agnosia: disorders of object recognition and what they tell us about normal vision. Cambridge, MA: MIT Press, 1990) of object, face and word recognition.     

DNA算法优化BP网络用于心律失常的识别

背景：目前对心律失常的诊断大多是由医生人工完成,费时费力,诊断结果依赖于医生的个人业务水平和责任心。心律失常的自动识别对于心脏病患者的救护和早期治疗具有非常重要的意义。 目的：实现临床心律失常的自动识别和诊断。 方法：首先从心电图中动态提取完整心律失常心拍形态,并采用离散余弦变换和反变换压缩数据;然后设计用于心律失常识别的BP神经网络,并用DNA算法优化该BP网络;最后用MIT/BIH心电数据库中心电图数据对DNA-BP网络进行检验。 结果与结论：对于5种心拍类型,包括正常、左束支阻滞、右束支阻滞、心室跳脱心搏及Paced 心搏,利用DNA-BP网络进行分类,实验达到了很好的识别效果,平均识别正确率达到99%。     

Detecting movement-related EEG change by wavelet decomposition-based neural networks trained with single thumb movement.

ObjectiveThe main goal of this study was to develop a real-time detection algorithm of movement-related EEG changes for the naïve subjects with a very small amount of training data. Such an algorithm is vital for the realization of brain–computer interface.MethodsThe target algorithm developed in this study was based on the wavelet decomposition neural network (WDNN). Surface Laplacian EEG was recorded at central cortical areas and processed with wavelet decomposition (WD) for feature extraction and neural network for pattern recognition. The new algorithm was compared with nother three methods, namely, threshold-based WD and short-time Fourier transform (STFT), and Fourier transform neural network (FTNN), for performance. The trainings of all algorithms were based, respectively, on the changes of μ and β rhythms before and after voluntary movements. In order to investigate whether WDNN could adapt to the nonstationarity of EEG or not, we also compared two training modes, namely, fixed and updated weight. The significances of the success rates were tested by ANOVA (analysis of variance) and verified by ROC (receiver operating characteristic) analysis.ResultsThe experimental data showed that (1) success rates of movement detection were acceptable even when the training set was reduced to a single trial data, (2) WDNN performed better than WD or STFT without optimized thresholds and (3) when weights were updated and thresholds were optimized, WDNN still performed better than WD, while FTNN had a marginal advantage over STFT.ConclusionsWe developed a detection algorithm based on WDNN with the training set being reduced to a single trial data. The overall performance of this algorithm was better than the conventional methods as such.Significanceμ wave suppression could be detected more precisely by the wavelet decomposition with neural network than the conventional algorithms such as STFT and WD. The size of training data could be reduced to a single trial and the success rates were up to 75–80%.