CiteULike: Group: vision-ng - library [278 articles]

Exponentiated Gradient Versus Gradient Descent for Linear Predictors

Jyrki Kivinen — 2007-09-04T08:15:47-00:00

No. UCSC-CRL-94-16. (1994)

We consider two algorithm for on-line prediction based on a linear model. The algorithms are the well-known gradient descent (GD) algorithm and a new algorithm, which we call EG Σ . They both maintain a weight vector using simple updates. For the GD algorithm, the update is based on subtracting the gradient of the squared error made on a prediction. The EG Σ algorithm uses the components of the gradient in the exponents of factors that are used in updating the weight vector...

Putting Objects in Perspective

Derek Hoiem — 2006-12-12T08:11:42-00:00

(2006)

Image understanding requires not only individually estimating elements of the visual world but also capturing the interplay among them. In this paper, we provide a framework for placing local object detection in the context of the overall 3D scene by modeling the interdependence of objects, surface orientations, and camera viewpoint.

Uncovering shared structures in multiclass classification

Yonatan Amit — 2007-08-20T17:35:22-00:00

(2007), pp. 17-24.

A dedicated generalized Procrustes algorithm for consensus molecular alignment

Jacques Commandeur — 2007-08-14T00:09:55-00:00

Journal of Chemometrics, Vol. 18, No. 1. (2004), pp. 37-42.

Recently the idea of using generalized Procrustes analysis for aligning sets of molecules was introduced using standard algorithms. In this paper it is shown that, by tailoring the algorithm to this specific problem, a great gain in computational speed and memory efficiency can be obtained, but even more importantly, by using rotations without reflection, changes in chirality of molecules can be prevented, which was not previously possible. Copyright © 2004 John Wiley & Sons, Ltd.

Nonlinear dimensionality reduction by semidefinite programming and kernel matrix factorization

W Weinberger — 2007-08-13T17:29:02-00:00

(2005)

We describe an algorithm for nonlinear dimensionality reduction based on semidefinite programming and kernel matrix factorization.

Learning a kernel matrix for nonlinear dimensionality reduction

Kilian Weinberger — 2006-10-22T16:10:42-00:00

(2004)

Unsupervised learning of image manifolds by semidefinite programming

KQ Weinberger — 2007-08-13T17:14:31-00:00

Computer Vision and Pattern Recognition, 2004. CVPR 2004. Proceedings of the 2004 IEEE Computer Society Conference on, Vol. 2 (2004), pp. II-988-II-995 Vol.2.

Can we detect low dimensional structure in high dimensional data sets of images and video? The problem of dimensionality reduction arises often in computer vision and pattern recognition. In this paper, we propose a new solution to this problem based on semidefinite programming. Our algorithm can be used to analyze high dimensional data that lies on or near a low dimensional manifold. It overcomes certain limitations of previous work in manifold learning, such as Isomap and locally linear embedding. We illustrate the algorithm on easily visualized examples of curves and surfaces, as well as on actual images of faces, handwritten digits, and solid objects.

Spectral Methods for Dimensionality Reduction

Lawrence Saul — 2007-08-09T08:19:11-00:00

(2005)

Online multiclass learning by interclass hypothesis sharing

Michael Fink — 2007-08-06T20:35:05-00:00

(2006), pp. 313-320.

A View of the EM Algorithm that Justifies Incremental, Sparse, and other Variants

R Neal — 2006-06-06T14:32:04-00:00

(1998)

. The EM algorithm performs maximum likelihood estimation for data in which some variables are unobserved. We present a function that resembles negative free energy and show that the M step maximizes this function with respect to the model parameters and the E step maximizes it with respect to the distribution over the unobserved variables. From this perspective, it is easy to justify an incremental variant of the EM algorithm in which the distribution for only one of the unobserved variables...

Rapid object detection using a boosted cascade of simple features

P Viola — 2005-05-16T15:51:34-00:00

(2001)

This paper describes a machine learning approach for visual object detection which is capable of processing images extremely rapidly and achieving high detection rates. This work is distinguished by three key contributions. The first is the introduction of a new image representation called the "Integral Image" which allows the features used by our detector to be computed very quickly. The second is a learning algorithm, based on AdaBoost, which selects a small number of critical visual features ...

Learning to detect objects in images via a sparse, part-based representation

S Agarwal — 2007-06-29T18:34:16-00:00

Pattern Analysis and Machine Intelligence, IEEE Transactions on, Vol. 26, No. 11. (2004), pp. 1475-1490.

We study the problem of detecting objects in still, gray-scale images. Our primary focus is the development of a learning-based approach to the problem that makes use of a sparse, part-based representation. A vocabulary of distinctive object parts is automatically constructed from a set of sample images of the object class of interest; images are then represented using parts from this vocabulary, together with spatial relations observed among the parts. Based on this representation, a learning algorithm is used to automatically learn to detect instances of the object class in new images. The approach can be applied to any object with distinguishable parts in a relatively fixed spatial configuration; it is evaluated here on difficult sets of real-world images containing side views of cars, and is seen to successfully detect objects in varying conditions amidst background clutter and mild occlusion. In evaluating object detection approaches, several important methodological issues arise that have not been satisfactorily addressed in the previous work. A secondary focus of this paper is to highlight these issues, and to develop rigorous evaluation standards for the object detection problem. A critical evaluation of our approach under the proposed standards is presented.

Old and new matrix algebra useful for statistics

T Minka — 2006-10-16T14:25:09-00:00

(1997)

This paper contains a large number of matrix identities which cannot be absorbed by mere reading. The reader is encouraged to take time and check each equation by hand and work out the examples. This is advanced material; see Searle (1982) for basic results. 1 Derivatives

A Hierarchical Community of Experts

Geoffrey Hinton — 2007-04-16T12:21:21-00:00

(1997)

We describe a directed acyclic graphical model that contains a hierarchy of linear units and a mechanism for dynamically selecting an appropriate subset of these units to model each observation. The non-linear selection mechanism is a hierarchy of binary units each of which gates the output of one of the linear units. There are no connections from linear units to binary units, so the generative model can be viewed as a logistic belief net (Neal 1992) which selects a skeleton linear model from...

Training Products of Experts by Maximizing Contrastive Likelihood

G Hinton — 2007-06-21T16:13:55-00:00

(1999)

It is possible to combine multiple probabilistic models of the same data by multiplying the probabilities together and then renormalizing. This is a very ecient way to model highdimensional data which simultaneously satises many dierent low-dimensional constraints because each individual expert model can focus on giving high probability to data vectors that satisfy just one of the constraints. Data vectors that satisfy this one constraint but violate other constraints will be ruled out...

Logistic Regression, AdaBoost and Bregman Distances

Michael Collins — 2007-06-04T22:12:37-00:00

(2000), pp. 158-169.

. We give a unified account of boosting and logistic regression in which each learning problem is cast in terms of optimization of Bregman distances. The striking similarity of the two problems in this framework allows us to design and analyze algorithms for both simultaneously, and to easily adapt algorithms designed for one problem to the other. For both problems, we give new algorithms and explain their potential advantages over existing methods. These algorithms can be divided into two...

Products of experts

G Hinton — 2007-05-16T15:18:17-00:00

(1999)

It is possible to combine multiple probabilistic models of the same data by multiplying the probabilities together and then renormalizing. This is a very ecient way to model high-dimensional data which simultaneously satises many dierent lowdimensional constraints. Each individual expert model can focus on giving high probability to data vectors that satisfy just one of the constraints. Data vectors that satisfy this one constraint but violate other constraints will be ruled out by their low...

Incorporating prior knowledge into boosting

R Schapire — 2007-06-04T05:35:26-00:00

(2002)

We describe a modification to the AdaBoost algorithm that permits the incorporation of prior human knowledge as a means of compensating for a shortage of training data. We give a convergence result for the algorithm.

Probabilistic principal component analysis

M Tipping — 2006-02-20T12:43:23-00:00

(1997)

Principal component analysis (PCA) is a ubiquitous technique for data analysis and processing, but one which is not based upon a probability model. In this paper we demonstrate how the principal axes of a set of observed data vectors may be determined through maximum-likelihood estimation of parameters in a latent variable model closely related to factor analysis. We consider the properties of the associated likelihood function, giving an EM algorithm for estimating the principal subspace...

Training Products of Experts by Minimizing Contrastive Divergence

Geoffrey Hinton — 2007-06-01T01:43:58-00:00

Neural Comp., Vol. 14, No. 8. (1 August 2002), pp. 1771-1800.

It is possible to combine multiple latent-variable models of the same data by multiplying their probability distributions together and then renormalizing. This way of combining individual "expert" models makes it hard to generate samples from the combined model but easy to infer the values of the latent variables of each expert, because the combination rule ensures that the latent variables of different experts are conditionally independent when given the data. A product of experts (PoE) is therefore an interesting candidate for a perceptual system in which rapid inference is vital and generation is unnecessary. Training a PoE by maximizing the likelihood of the data is difficult because it is hard even to approximate the derivatives of the renormalization term in the combination rule. Fortunately, a PoE can be trained using a different objective function called "contrastive divergence" whose derivatives with regard to the parameters can be approximated accurately and efficiently. Examples are presented of contrastive divergence learning using several types of expert on several types of data.

Learning Probabilistic Relational Models

Lise Getoor — 2005-12-30T17:22:38-00:00

Lecture Notes in Computer Science, Vol. 1864 (2000), pp. 322-??.

Most real-world data is stored in relational form. In contrast, most statistical learning methods, e.g., Bayesian network learning, work only with "flat" data representations, forcing us to convert our data into a form that loses much of the relational structure. The recently introduced framework of probabilistic relational models (PRMs) allow us to represent much richer dependency structures, involving multiple entities and the relations between them; they allow the properties of an...

End stopping in V1 is sensitive to contrast

Arash Yazdanbakhsh — 2006-04-25T16:30:23-00:00

Nature Neuroscience, Vol. 9, No. 5. (23 April 2006), pp. 697-702.

Improving text classification by shrinkage in a hierarchy of classes

A Mccallum — 2006-10-16T16:19:28-00:00

(1998)

When documents are organized in a large number of topic categories, the categories are often arranged in a hierarchy. The U.S. patent database and Yahoo are two examples.

Learning Multiple Feature Representations from Natural Image Sequences

Wolfgang Einhã¤user — 2007-01-29T16:59:43-00:00

: Artificial Neural Networks - ICANN 2002: International Conference, Madrid, Spain, August 28-30, 2002. Proceedings (2002), 21.

Contrast gain control in the cat visual cortex.

I Ohzawa — 2007-01-29T16:46:51-00:00

Nature, Vol. 298, No. 5871. (15 July 1982), pp. 266-268.

The eye functions effectively over an enormous range of ambient illumination, because retinal sensitivity can be adapted to prevailing light levels. Higher order neurones in the visual pathway are presumably more concerned with relative changes in illumination, that is, contrast, because a great deal of information concerning absolute light level is processed at the retinal level. It would therefore be of considerable functional value if cells in the visual cortex could adapt their response levels to a steady-state ambient contrast, in a manner analogous to the sensitivity control mechanism of the retina. We have examined here the idea that adaptation of neurones in the visual cortex to ambient contrast is similar to adaptation in the retina to ambient illumination. The experiments were performed by measuring contrast response functions (response amplitude as a function of contrast) of striate neurones, while systematically adapting them to different contrast levels. Our results show that, for the majority of cortical neurones, response-contrast curves are laterally shifted along a log-contrast axis so that the effective domains of neurones are adjusted to match prevailing contrast levels. This contrast gain control mechanism, which was not observed for lateral geniculate (LGN) fibres, must be of prime importance to visual function.

Simple-cell-like receptive fields maximize temporal coherence in natural video.

J Hurri — 2007-01-29T16:44:45-00:00

Neural Comput, Vol. 15, No. 3. (March 2003), pp. 663-691.

Recently, statistical models of natural images have shown the emergence of several properties of the visual cortex. Most models have considered the nongaussian properties of static image patches, leading to sparse coding or independent component analysis. Here we consider the basic time dependencies of image sequences instead of their nongaussianity. We show that simple-cell-type receptive fields emerge when temporal response strength correlation is maximized for natural image sequences. Thus, temporal response strength correlation, which is a nonlinear measure of temporal coherence, provides an alternative to sparseness in modeling simple-cell receptive field properties. Our results also suggest an interpretation of simple cells in terms of invariant coding principles, which have previously been used to explain complex-cell receptive fields.

Isolation of relevant visual features from random stimuli for cortical complex cells.

J Touryan — 2007-01-29T16:41:01-00:00

J Neurosci, Vol. 22, No. 24. (15 December 2002), pp. 10811-10818.

A crucial step in understanding the function of a neural circuit in visual processing is to know what stimulus features are represented in the spiking activity of the neurons. For neurons with complex, nonlinear response properties, characterization of feature representation requires measurement of their responses to a large ensemble of visual stimuli and an analysis technique that allows identification of relevant features in the stimuli. In the present study, we recorded the responses of complex cells in the primary visual cortex of the cat to spatiotemporal random-bar stimuli and applied spike-triggered correlation analysis of the stimulus ensemble. For each complex cell, we were able to isolate a small number of relevant features from a large number of null features in the random-bar stimuli. Using these features as visual stimuli, we found that each relevant feature excited the neuron effectively in isolation and contributed to the response additively when combined with other features. In contrast, the null features evoked little or no response in isolation and divisively suppressed the responses to relevant features. Thus, for each cortical complex cell, visual inputs can be decomposed into two distinct types of features (relevant and null), and additive and divisive interactions between these features may constitute the basic operations in visual cortical processing.

How are complex cell properties adapted to the statistics of natural stimuli?

KP Körding — 2005-11-27T12:08:59-00:00

J Neurophysiol, Vol. 91, No. 1. (January 2004), pp. 206-212.

Sensory areas should be adapted to the properties of their natural stimuli. What are the underlying rules that match the properties of complex cells in primary visual cortex to their natural stimuli? To address this issue, we sampled movies from a camera carried by a freely moving cat, capturing the dynamics of image motion as the animal explores an outdoor environment. We use these movie sequences as input to simulated neurons. Following the intuition that many meaningful high-level variables, e.g., identities of visible objects, do not change rapidly in natural visual stimuli, we adapt the neurons to exhibit firing rates that are stable over time. We find that simulated neurons, which have optimally stable activity, display many properties that are observed for cortical complex cells. Their response is invariant with respect to stimulus translation and reversal of contrast polarity. Furthermore, spatial frequency selectivity and the aspect ratio of the receptive field quantitatively match the experimentally observed characteristics of complex cells. Hence, the population of complex cells in the primary visual cortex can be described as forming an optimally stable representation of natural stimuli.

Computational subunits of visual cortical neurons revealed by artificial neural networks.

B Lau — 2006-02-17T22:18:20-00:00

Proc Natl Acad Sci U S A, Vol. 99, No. 13. (25 June 2002), pp. 8974-8979.

A crucial step toward understanding visual processing is to obtain a comprehensive description of the relationship between visual stimuli and neuronal responses. Many neurons in the visual cortex exhibit nonlinear responses, making it difficult to characterize their stimulus-response relationships. Here, we recorded the responses of primary visual cortical neurons of the cat to spatiotemporal random-bar stimuli and trained artificial neural networks to predict the response of each neuron. The random initial connections in the networks consistently converged to regular patterns. Analyses of these connection patterns showed that the response of each complex cell to the random-bar stimuli could be well approximated by the sum of a small number of subunits resembling simple cells. The direction selectivity of each complex cell measured with drifting gratings was also well predicted by the combination of these subunits, indicating the generality of the model. These results are consistent with a simple functional model for complex cells and demonstrate the usefulness of the neural network method for revealing the stimulus-response transformations of nonlinear neurons.

Nonlinear and extra-classical receptive field properties and the statistics of natural scenes.

C Zetzsche — 2007-01-29T16:27:24-00:00

Network, Vol. 12, No. 3. (August 2001), pp. 331-350.

The neural mechanisms of early vision can be explained in terms of an information-theoretic optimization of the neural processing with respect to the statistical properties of the natural environment. Recent applications of this approach have been successful in the prediction of the linear filtering properties of ganglion cells and simple cells, but the relations between the environmental statistics and cortical nonlinearities, like those of end-stopped or complex cells, are not yet fully understood. Here we present extensions of our previous investigations of the exploitation of higher-order statistics by nonlinear neurons. We use multivariate wavelet statistics to demonstrate that a strictly linear processing would inevitably leave substantial statistical dependencies between the outputs of the units. We then consider how the basic nonlinearities of cortical neurons--gain control and ON/OFF half-wave rectification--can exploit these higher-order statistical dependencies. We first show that gain control provides an adaptation to the polar separability of the multivariate probability density function (PDF), and, together with an output nonlinearity, enables an overcomplete sparse coding. We then consider how the remaining higher-order dependencies between different units can be exploited by a combination of basic ON/OFF point nonlinearities and subsequent weighted linear combinations. We consider two statistical optimization schemes for the computation of the optimal weights: principal component analysis (PCA) and independent component analysis (ICA). Since the intermediate nonlinearities transform some of the higher-order dependencies into second-order dependencies even the basic PCA approach is able to exploit part of the redundancies. ICA ignores this second-order structure, but can exploit higher-order dependencies. Both schemes yield a variety of nonlinear units which comprise the typical nonlinear processing properties, such as end-stopping, side-stopping, complex-cell properties and extra-classical receptive field properties, but the 'ideal' complex cells seem only to occur with PCA. Thus, a combination of ON/OFF nonlinearities with an integrated PCA-ICA strategy seems necessary to exploit the statistical properties of natural images.

Quadratic forms in natural images.

W Hashimoto — 2007-01-29T16:16:10-00:00

Network, Vol. 14, No. 4. (November 2003), pp. 765-788.

Several studies have succeeded in correlating natural image statistics with receptive field properties of neurons in the primary visual cortex. If we determine the parameters of linear transformations that make their output values as independent as possible when input data are natural images, we obtain parameter values that correspond to simple cell characteristics. It was also proved that, by making output values as temporally coherent as possible, simple cell characteristics also emerge. However, complex cell properties have not been fully explained by previous studies of natural image statistics. In this study, we examine whether we could reproduce complex cell properties by determining the parameters of two-layer networks that make their outputs as independent and sparse as possible or as temporally coherent as possible. Input-output functions of two-layer networks correspond to quadratic forms and they form a class of functions that includes complex cell responses and many other functions. Therefore, we employed two-layer networks as a framework for discussing complex cell properties as in previous studies. By maximizing the independence and sparseness of output values of two-layer networks without considering the temporal structure of input images, squared responses of simple cells are obtained and complex cell properties are not reproduced. On the other hand, by maximizing the temporal coherence of output, we obtain complex cell properties among other kinds of input-output functions. In previous studies, the measure of temporal coherence was the squared difference between the responses to two consecutive input images. We obtain two-layer networks that minimize this measure and show that some of them exhibit properties of complex cells but not clearly. We propose the sparseness of difference between responses to two consecutive inputs as an alternative measure of temporal coherence. We formulate an algorithm to maximize the sparseness of difference and show that complex cell properties emerge more clearly.

Expected sample moments of concomitants of selected order statistics

D Arnold — 2007-01-25T17:23:41-00:00

(2002)

In this paper, the task of determining expected values of sample moments, where the sample members have been selected based on noisy information, is considered. Exact expressions for expected values of sums of products of concomitants of selected order statistics are derived. Then, using Edgeworth and Cornish-Fisher approximations, explicit results that depend on coefficients that can be determined numerically are obtained. While the results are exact only for normal populations, it is...

Errors in the estimation of the variance: implications for multiple-probability fluctuation analysis.

C Saviane — 2007-01-09T11:51:38-00:00

J Neurosci Methods, Vol. 153, No. 2. (15 June 2006), pp. 250-260.

Synapses play a crucial role in information processing in the brain. Amplitude fluctuations of synaptic responses can be used to extract information about the mechanisms underlying synaptic transmission and its modulation. In particular, multiple-probability fluctuation analysis can be used to estimate the number of functional release sites, the mean probability of release and the amplitude of the mean quantal response from fits of the relationship between the variance and mean amplitude of postsynaptic responses, recorded at different probabilities. To determine these quantal parameters, calculate their uncertainties and the goodness-of-fit of the model, it is important to weight the contribution of each data point in the fitting procedure. We therefore investigated the errors associated with measuring the variance by determining the best estimators of the variance of the variance and have used simulations of synaptic transmission to test their accuracy and reliability under different experimental conditions. For central synapses, which generally have a low number of release sites, the amplitude distribution of synaptic responses is not normal, thus the use of a theoretical variance of the variance based on the normal assumption is not a good approximation. However, appropriate estimators can be derived for the population and for limited sample sizes using a more general expression that involves higher moments and introducing unbiased estimators based on the h-statistics. Our results are likely to be relevant for various applications of fluctuation analysis when few channels or release sites are present.

On the Mathematical Expectation of the Moments of Frequency Distributions

Al — 2007-01-25T17:16:58-00:00

Estimation of the accuracy of mean and variance of correlated data

PMT Broersen — 2007-01-25T09:00:44-00:00

Instrumentation and Measurement Technology Conference, 1998. IMTC/98. Conference Proceedings. IEEE, Vol. 1 (1998), pp. 36-41 vol.1.

Monte Carlo simulations are an important tool in computational physics or statistical mechanics. Physical constants or properties are found as the mean or the variance of successive states of simulated systems. A new method to determine the statistical accuracy, of the estimated means and variances is described. It uses the parameters of an automatically selected time series model. That time series model gives an optimal description of the spectral density and of the correlation structure of correlated data which are considered as stationary or in equilibrium. The resulting accuracy estimates are close to the Cramer-Rao bound for data where the correlation is determined by a single time constant

Emergence of movement sensitive neurons' properties by learning a sparse code of natural moving images

Rafal Bogacz — 2007-01-24T00:27:46-00:00

(2001)

Shift-invariant recognition by the conjunction of basic invariant patterns

S Soloviev — 2007-01-18T01:50:55-00:00

(1997)

An approach to shift-invariant pattern recognition is presented, based on the conjunction of Basic Invariant Patterns (BIPs). In this approach, the invariance to complex shapes is built from the invariance to simple basic patterns. Each basic pattern is recognized at any given location by localized detectors for this pattern. Complex shapes are then treated as conjunctions of the basic patterns. To test the ideas, a simple shift-invariant system was implemented with input shapes being line...

Computation of pattern invariance in brain-like structures.

S Ullman — 2007-01-18T01:50:03-00:00

Neural Netw, Vol. 12, No. 7-8. (October 1999), pp. 1021-1036.

A fundamental capacity of the perceptual systems and the brain in general is to deal with the novel and the unexpected. In vision, we can effortlessly recognize a familiar object under novel viewing conditions, or recognize a new object as a member of a familiar class, such as a house, a face, or a car. This ability to generalize and deal efficiently with novel stimuli has long been considered a challenging example of brain-like computation that proved extremely difficult to replicate in artificial systems. In this paper we present an approach to generalization and invariant recognition. We focus our discussion on the problem of invariance to position in the visual field, but also sketch how similar principles could apply to other domains.The approach is based on the use of a large repertoire of partial generalizations that are built upon past experience. In the case of shift invariance, visual patterns are described as the conjunction of multiple overlapping image fragments. The invariance to the more primitive fragments is built into the system by past experience. Shift invariance of complex shapes is obtained from the invariance of their constituent fragments. We study by simulations aspects of this shift invariance method and then consider its extensions to invariant perception and classification by brain-like structures.

Learning invariance from transformation sequences

Peter F&\#246;ldi&\#225;k — 2007-01-18T01:48:35-00:00

Neural Comput., Vol. 3, No. 2. (1991), pp. 194-200.

Greedy Layer-Wise Training of Deep Networks

Yoshua Bengio — 2007-01-18T01:28:32-00:00

Fast and Robust Fixed-Point Algorithms for Independent Component Analysis

A Hyvärinen — 2007-01-18T01:19:22-00:00

IEEE Transactions on Neural Networks, Vol. 10, No. 3. (1999), pp. 626-634.

Independent component analysis (ICA) is a statistical method for transforming an observed multidimensional random vector into components that are statistically as independent from each other as possible. In this paper, we use a combination of two different approaches for linear ICA: Comon's information-theoretic approach and the projection pursuit approach. Using maximum entropy approximations of differential entropy, we introduce a family of new contrast (objective) functions for ICA. These...

Efficient Learning of Sparse Representations with an Energy-Based Model

Marc' Ranzato — 2007-01-18T01:15:34-00:00

(2006)

Slow feature analysis yields a rich repertoire of complex cell properties.

P Berkes — 2005-10-06T15:38:50-00:00

J Vis, Vol. 5, No. 6. (20 July 2005), pp. 579-602.

In this study we investigate temporal slowness as a learning principle for receptive fields using slow feature analysis, a new algorithm to determine functions that extract slowly varying signals from the input data. We find a good qualitative and quantitative match between the set of learned functions trained on image sequences and the population of complex cells in the primary visual cortex (V1). The functions show many properties found also experimentally in complex cells, such as direction selectivity, non-orthogonal inhibition, end-inhibition, and side-inhibition. Our results demonstrate that a single unsupervised learning principle can account for such a rich repertoire of receptive field properties.

Slow feature analysis: unsupervised learning of invariances.

L Wiskott — 2005-10-17T15:02:28-00:00

Neural Comput, Vol. 14, No. 4. (April 2002), pp. 715-770.

Invariant features of temporally varying signals are useful for analysis and classification. Slow feature analysis (SFA) is a new method for learning invariant or slowly varying features from a vectorial input signal. It is based on a nonlinear expansion of the input signal and application of principal component analysis to this expanded signal and its time derivative. It is guaranteed to find the optimal solution within a family of functions directly and can learn to extract a large number of decorrelated features, which are ordered by their degree of invariance. SFA can be applied hierarchically to process high-dimensional input signals and extract complex features. SFA is applied first to complex cell tuning properties based on simple cell output, including disparity and motion. Then more complicated input-output functions are learned by repeated application of SFA. Finally, a hierarchical network of SFA modules is presented as a simple model of the visual system. The same unstructured network can learn translation, size, rotation, contrast, or, to a lesser degree, illumination invariance for one-dimensional objects, depending on only the training stimulus. Surprisingly, only a few training objects suffice to achieve good generalization to new objects. The generated representation is suitable for object recognition. Performance degrades if the network is trained to learn multiple invariances simultaneously.

Visual speech speeds up the neural processing of auditory speech.

V van Wassenhove — 2005-01-27T03:55:27-00:00

Proc Natl Acad Sci U S A, Vol. 102, No. 4. (25 January 2005), pp. 1181-1186.

Synchronous presentation of stimuli to the auditory and visual systems can modify the formation of a percept in either modality. For example, perception of auditory speech is improved when the speaker's facial articulatory movements are visible. Neural convergence onto multisensory sites exhibiting supra-additivity has been proposed as the principal mechanism for integration. Recent findings, however, have suggested that putative sensory-specific cortices are responsive to inputs presented through a different modality. Consequently, when and where audiovisual representations emerge remain unsettled. In combined psychophysical and electroencephalography experiments we show that visual speech speeds up the cortical processing of auditory signals early (within 100 ms of signal onset). The auditory-visual interaction is reflected as an articulator-specific temporal facilitation (as well as a nonspecific amplitude reduction). The latency facilitation systematically depends on the degree to which the visual signal predicts possible auditory targets. The observed auditory-visual data support the view that there exist abstract internal representations that constrain the analysis of subsequent speech inputs. This is evidence for the existence of an "analysis-by-synthesis" mechanism in auditory-visual speech perception.

Partially observed values

T Raiko — 2007-01-07T04:36:10-00:00

Neural Networks, 2004. Proceedings. 2004 IEEE International Joint Conference on, Vol. 4 (2004), pp. 2825-2830 vol.4.

It is common to have both observed and missing values in data. This paper concentrates on the case where a value can be somewhere between those two ends, partially observed and partially missing. To achieve that, a method of using evidence nodes in a Bayesian network is studied. Different ways of handling inaccuracies are discussed in examples and the proposed approach is justified in the experiments with real image data. Also, a justification is given for the standard preprocessing step of adding a tiny amount of noise to the data, when a continuous valued model is used for discrete-valued data.

Death, Hope, and Sex: Life-History Theory and the Development of Reproductive Strategies [and Comments and Reply]

James Chisholm — 2006-12-05T12:55:43-00:00

Current Anthropology, Vol. 34, No. 1. (1993), pp. 1-24.

Parental investment in male and female offspring in polygynous mammals

TH Clutton-Brock — 2006-12-05T12:43:48-00:00

Nature, Vol. 289, No. 5797. (5 February 1981), pp. 487-489.

Parent-Offspring Conflict

Robert Trivers — 2006-12-05T12:40:40-00:00

Amer. Zool., Vol. 14, No. 1. (1 January 1974), pp. 249-264.

When parent-offspring relations in sexually reproducing species are viewed from the standpoint of the offspring as well as the parent, conflict is seen to be an expected feature of such relations. In particular, parent and offspring are expected to disagree over how long the period of parental investment should last, over the amount of parental investment that should be given, and over the altruistic and egoistic tendencies of the offspring as these tendencies affect other relatives. In addition, under certain conditions parents and offspring are expected to disagree over the preferred sex of the potential offspring. In general, parent-offspring conflict is expected to increase during the period of parental care, and offspring are expected to employ psychological weapons in order to compete with their parents. Detailed data on mother-offspring relations in mammals are consistent with the arguments presented. Conflict in some species, including the human species, is expected to extend to the adult reproductive role of the offspring: under certain conditions parents are expected to attempt to mold an offspring, against its better interests, into a permanent nonreproductive. 10.1093/icb/14.1.249

Male parental care, differential parental investment by females and sexual selection

AP Moller — 2006-12-05T12:27:12-00:00

pp. 1507-1515.

Males play a variable parental role in reproduction, ranging from no male parental care to extensive male care. Females may acquire either direct or indirect fitness benefits from their mate choice, and direct fitness benefits include male parental care. Theoreticians have traditionally emphasized direct fitness benefits to females in species with extensive male parental care. We review the literature and show extensive variation in the patterns of male care, related to the attractiveness of males to females. At one extreme of this continuum, females invest differentially in parental care, investing more when paired with attractive males. The costs of female parental care and other aspects of parental investment may be balanced by benefits in terms of more attractive sons and/or more viable offspring. At the other extreme, in species with extensive direct fitness benefits, males with preferred sexual phenotypes provide the largest relative share of parental care. A comparative study of birds revealed that the extent of the differential female parental investment was directly related to the frequency of extra-pair paternity. Since extra-pair paternity may arise mainly as a consequence of female choice for indirect fitness benefits, this result supports our prediction that differential parental investment is prevalent in species where females benefit indirectly from their mate choice. The consequences for sexual selection theory of these patterns of male care in relation to male attractiveness are emphasized. Copyright 1998 The Association for the Study of Animal Behaviour.