Adaptive pattern classification and universal recoding: II. Feedback,expectation, olfaction,illusions

Part I of this paper describes a model for the parallel development and adult coding of neural feature detectors. It shows how any set of arbitrary spatial patterns can be recoded, or transformed, into any other spatial patterns (universal recoding), if there are sufficiently many cells in the network's cortex. This code is, however, unstable through time if arbitrarily many patterns can perturb a fixed number of cortical cells. This paper shows how to stabilize the code in the general case using feedback between cellular sites. A biochemically defined critical period is not necessary to stabilize the code, nor is it sufficient to ensure useful coding properties.We ask how short term memory can be reset in response to temporal sequences of spatial patterns. This leads to a context-dependent code in which no feature detector need uniquely characterize an input pattern; yet unique classification by the pattern of activity across feature detectors is possible. This property uses learned expectation mechanisms whereby unexpected patterns are temporarily suppressed and/or activate nonspecific arousal. The simplest case describes reciprocal interactions via trainable synaptic pathways (long term memory traces) between two recurrent on-center off-surround networks undergoing mass action (shunting) interactions. This unit can establish an adaptive resonance, or reverberation, between two regions if their coded patterns match, and can suppress the reverberation if their patterns do not match. This concept yields a model of olfactory coding within the olfactory bulb and prepyriform cortex. The resonance idea also includes the establishment of reverberation between conditioned reinforcers and generators of contingent negative variation if presently avialable sensory cues are compatible with the network's drive requirements at that time; and a search and lock mechanism whereby the disparity between two patterns can be minimized and the minimal disparity images locked into position. Stabilizing the code uses attentional mechanisms, in particular nonspecific arousal as a tuning and search device. We suggest that arousal is gated by a chemical transmitter system—for example, norepinephrine—whose relative states of accumulation at antagonistic pairs of on-cells and off-cells through time can shift the spatial pattern of STM activity across a field of feature detectors. For example, a sudden arousal increment in response to an un-expected pattern can reverse, or rebound, these relative activities, thereby suppressing incorrectly classified populations. The rebound mechanism has formal properties analogous to negative afterimages and spatial frequency adaptation.Supported in part by the Advanced Research Projects Agency under ONR Contract No. N00014-76-C-0185     

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

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.     

Oscillatory neural network for pattern recognition: trajectory based classification and supervised learning

Computer algorithms that match human performance in recognizing written text or spoken conversation remain elusive. The reasons why the human brain far exceeds any existing recognition scheme to date in the ability to generalize and to extract invariant characteristics relevant to category matching are not clear. However, it has been postulated that the dynamic distribution of brain activity (spatiotemporal activation patterns) is the mechanism by which stimuli are encoded and matched to categories. This research focuses on supervised learning using a trajectory based distance metric for category discrimination in an oscillatory neural network model. Classification is accomplished using a trajectory based distance metric. Since the distance metric is differentiable, a supervised learning algorithm based on gradient descent is demonstrated. Classification of spatiotemporal frequency transitions and their relation to a priori assessed categories is shown along with the improved classification results after supervised training. The results indicate that this spatiotemporal representation of stimuli and the associated distance metric is useful for simple pattern recognition tasks and that supervised learning improves classification results.     

Dental development in South African australopithecines. Part I: Problems of pattern and chronology

It is well known that humans take about twice as long as apes to mature. The traditional view that such delayed maturation was already present in australopithecines has been called into question during the past several years. We have approached this problem by looking at patterns of dental development in gracile and robust australopithecines from South Africa and comparing them to patterns found in extant humans and apes. We have employed both 2 and 3 dimensional computed tomography in our research. The dental growth patterns in these two australopithecine morphs differ, particularly in M1/I1 development. The robust australopithecines are more humanlike and the gracile australopithecines more apelike in this feature (“humanlike” and “apelike” are not used in any taxonomic sense). Pattern and chronology of dental development must be considered separately. Several major problem areas for future research are identified, most of which revolve around the issue of intra- versus interspecific variation.     

A combinational feature selection and ensemble neural network method for classification of gene expression data

Background  

Microarray experiments are becoming a powerful tool for clinical diagnosis, as they have the potential to discover gene expression patterns that are characteristic for a particular disease. To date, this problem has received most attention in the context of cancer research, especially in tumor classification. Various feature selection methods and classifier design strategies also have been generally used and compared. However, most published articles on tumor classification have applied a certain technique to a certain dataset, and recently several researchers compared these techniques based on several public datasets. But, it has been verified that differently selected features reflect different aspects of the dataset and some selected features can obtain better solutions on some certain problems. At the same time, faced with a large amount of microarray data with little knowledge, it is difficult to find the intrinsic characteristics using traditional methods. In this paper, we attempt to introduce a combinational feature selection method in conjunction with ensemble neural networks to generally improve the accuracy and robustness of sample classification.     

Temporal coding: the relevance of classical activity, its relation to pattern frequency bands, and a remark on recoding of excitatory drive into phase shifts

We consider an oscillatory network model that is obtained as complex-valued generalization of the classical Cohen-Grossberg-Hopfield (CGH) model. Apart from a synchronizing mechanism, a stronger and/or more coherent input to a unit in the network implies a higher phase velocity of this unit. This constitutes the desynchronizing mechanism, referred to as acceleration. The units' activity of the classical model translates into the amplitudes of the phase model oscillators. This allows to associate classical and temporal coding with amplitude and phase dynamics, respectively. We discuss how the two dynamics act together to achieve the unambiguous pattern recognition that avoids the superposition problem. With respect to coherence, dominating patterns may take coherent states also if only a subset of its units is on-state. The competition for coherence, introduced by acceleration, realizes a kind of feature counting that identifies the dominating pattern as the pattern with the most on-state units. This dominating but possibly only partially active pattern may take a coherent state with a frequency level that is related to the number of on-state units. We also speculate on neurophysiological findings, related to observed phase differences between optimally and suboptimally activated neurons, that may indicate the presence of acceleration.     

Evolution of macromolecules: I. Dual coding systems

The evolution of proteins determined by two independently mutable coding mechanisms (e.g., one in which nucleic acids operate with unit coding ratio) has been analyzed kinetically in two ways. The first presumes a system of mutating and reproducing proteins; the dependent variables are the numbers of the various kinds of proteins—wild types, and mutants obtained by mutations in one or another of the two coding mechanisms. The second approach deals with kinds rather than numbers of proteins; the reproductive element in the evolving system is dealt with by assuming a specific rate of extinction for members of each protein class, due to the occurrence of lethal mutations in the proteins themselves or in other proteins in the organisms that carry them. If the two kinds of mutants occur at different rates, it is shown in both treatments that time will not necessarily extinguish the initial advantage of one of them—that is, the notion that the slower class will eventually occur often enough to produce a random distribution of the two classes after long periods of evolution is not in general true. The effects of mutation rate, reproduction rate, and extinction rate on the distribution of the various protein classes are analyzed. Contribution No. 666 from the Division of Basic Health Sciences.     

On the development of feature detectors in the visual cortex with applications to learning and reaction-diffusion systems

Developmental mechanisms for tuning of visual cortex are derived from adult learning mechanisms: an adaptational property of shunting on-center off-surround networks that prevents saturation of parallel processed patterns at high input intensities, a contrast enhancement and short-term memory mechanism, and plastic synaptic strengths that compute a time average of presynaptic signals and postsynaptic activities and multiplicatively gate signals. The mechanisms can generate fields of feature detectors; e.g., line or picture detectors. A developing hierarchy of such fields can be synthesized in which successive critical developmental periods are triggered as a dynamic equilibrium is established between shortterm memory and long-term memory at each stage. Shunting adaptation can account for some data on spatial frequency adaptation. Shunting network properties resemble properties of certain reaction-diffusion systems that have been used to model developmental data in various species; e.g., Hydra, Xenopus retina, slime molds. For example, positional information due to regulation in reaction-diffusion systems is analogous to constancies due to network adaptation, firing of a developmental gradient is analogous to contrast enhancement, and maintenance of a pattern of morphogens is analogous to short-term memory.Supported in part by the Advanced Research Projects Agency (DAHC15-73-C-0320) administered by Computer Corporation America.     

Computer analysis of cervical cells. Automatic feature extraction and classification

A prescreening instrument for cervical smears using computerized image processing and pattern recognition techniques requires that single cells in the specimen can be automatically isolated and analyzed. This paper describes a dual wavelength method for automatic isolation of the cytoplasm and nuclei of cells. Density-oriented, shape-oriented and texture-oriented parameters were calculated and evaluated for more than 600 cells. It is shown that the computer can be taught to distinguish between normal and atypical cells with an accuracy of ca. 97%, while human classification reproducibility is ca. 95%. In addition, an attempt to assign a measure of atypia to individual cells is described.     

A proposal for the classification and evaluation of fall detectors

Falls affect, each year, tens of million of elderly people throughout the world. It can have immediate lethal consequences but also causes many disabling fractures and dramatic psychological consequences which reduce the independence of the person. Falls in the elderly is thus a major public health problem. The “early” detection of fall consequently raises the interest of searchers, as most of elderly fallers cannot return to a standing position on their own following a fall. It is also an interesting scientific problem because it is an ill-defined process. The goals of this study were to classify various approaches used to detect the fall and to point out the difficulty to compare the results of these studies, as there is currently no common evaluation benchmark.     

Developing classification in action: I. Human infants

Human infants’ developing manipulatory transformations involved in classifying objects from ages 6 to 24 months were investigated. Infants’ manipulations develop from predominantly serial one-at-a-time acts with one object to predominantly parallel two-at-a-time acts with two objects. This shift is marked by increasingly overt transformational consequences for the objects manipulated. When infants construct parallel transformations they are initially different. With age they are increasingly identical or reciprocal. Also during this age period, as the number of objects manipulated at the same time increases, so does the frequency with which infants coordinate them. At the same time, the kind of objects infants manipulate simultaneously changes. Six-month-olds manipulate different objects when acting on more than one object at a time. By age 12 months, infants switch to manipulating identical objects at the same time, indicating that they are beginning to construct identity classes. Since this development occurs about a half year before human infants develop any substantial naming behavior, the origins of classification cannot depend on this linguistic development     

Biological and bionic hands: natural neural coding and artificial perception

The first decade and a half of the twenty-first century brought about two major innovations in neuroprosthetics: the development of anthropomorphic robotic limbs that replicate much of the function of a native human arm and the refinement of algorithms that decode intended movements from brain activity. However, skilled manipulation of objects requires somatosensory feedback, for which vision is a poor substitute. For upper-limb neuroprostheses to be clinically viable, they must therefore provide for the restoration of touch and proprioception. In this review, I discuss efforts to elicit meaningful tactile sensations through stimulation of neurons in somatosensory cortex. I focus on biomimetic approaches to sensory restoration, which leverage our current understanding about how information about grasped objects is encoded in the brain of intact individuals. I argue that not only can sensory neuroscience inform the development of sensory neuroprostheses, but also that the converse is true: stimulating the brain offers an exceptional opportunity to causally interrogate neural circuits and test hypotheses about natural neural coding.     

Learning and coding of concepts in neural networks

Our environment consists of virtually an infinite number of scenarios in which we have to function. In order to respond properly to an incoming stimulus, the brain has first to analyze it, and to find out the basic familiar elements that are part of it. In other words, by using a library which contains a relatively small number of basic concepts, the brain analyzes the multitude of incoming events. Some of those basic concepts are innate, but many of them must be learned, in order to accommodate for the arbitrary environment around us. A classifying box is defined as the neural network that finds out the familiar concepts that are present in an incoming stimulus. Models for classifying boxes are introduced, and possible mechanisms by which they may establish their libraries of concepts are suggested, and then compared and evaluated by computer simulations.