Marc M. Van Hulle

Shape and Spatial Distribution of Receptive Fields and Antagonistic Motion Surrounds in the Middle Temporal Area (V5) of the Macaque

The spatial organization of receptive fields in the middle temporal (MT) area of anaesthetized and paralysed macaque monkeys was studied. In all, 288 neurons were successfully recorded. The size and shape of the receptive field (RF) was mapped with small patches of translating random dots and the resulting data were fitted with a generalized Gaussian. Results show that the RF area increases with eccentricity, and is larger in lamina 5 than in other layers. Most of these RFs are elongated, and the axis of elongation tends to be orthogonal to the preferred direction of motion. The direction selectivity is maintained in all positions in the RF, but layer 5 cells are less direction‐selective than cells in other layers. In a second series of experiments, radial dimensions of the classical RF and the antagonistic surround were estimated from area summation tests. These data were fitted with the difference of the integrals of two Gaussians. Surrounds were weakest in layer 4 and strongest in layer 2. Optimal stimulus diameters, also estimated from the area summation curve, were larger in the infragranular layers than in the other layers. The maximum sensitivity of the surround was clearly displaced from the classical RF (CRF) centre, indicating that the surround is not concentric with the CRF. This radial offset and the extent of the surround were largest in layers 2 and 5 and smallest in 3a. The extent of the surround half‐height equalled, on average, 3–4 times that of the CRF. These results suggest that antagonistic surrounds are constructed in MT, probably through horizontal connections, and that a strong vertical organization exists in area MT, as has been shown for V1.

Edgeworth Approximation of Multivariate Differential Entropy

We develop the general, multivariate case of the Edgeworth approximation of differential entropy and show that it can be more accurate than the nearest-neighbor method in the multivariate case and that it scales better with sample size. Furthermore, we introduce mutual information estimation as an application.

Kernel-based equiprobabilistic topographic map formation

We introduce a new unsupervised competitive learning rule, the kernel-based maximum entropy learning rule (kMER), which performs equiprobabilistic topographic map formation in regular, fixed-topology lattices, for use with nonparametric density estimation as well as nonparametric regression analysis. The receptive fields of the formal neurons are overlapping radially symmetric kernels, compatible with radial basis functions (RBFs); but unlike other learning schemes, the radii of these kernels do not have to be chosen in an ad hoc manner: the radii are adapted to the local input density, together with the weight vectors that define the kernel centers, so as to produce maps of which the neurons have an equal probability to be active (equiprobabilistic maps). Both an online and a batch version of the learning rule are introduced, which are applied to nonparametric density estimation and regression, respectively. The application envisaged is blind source separation (BSS) from nonlinear, noisy mixtures.

The formation of topographic maps that maximize the average mutual information of the output responses to noiseless input signals

This article introduces an extremely simple and local learning rule for to pographic map formation. The rule, called the maximum entropy learning rule (MER), maximizes the unconditional entropy of the maps output for any type of input distribution. The aim of this article is to show that MER is a viable strategy for building topographic maps that maximize the average mutual information of the output responses to noiseless input signals when only input noise and noise-added input signals are available.

ENHANCING THE YIELD OF HIGH-DENSITY ELECTRODE ARRAYS THROUGH AUTOMATED ELECTRODE SELECTION

Recently developed CMOS-based microprobes contain hundreds of electrodes on a single shaft with inter-electrode distances as small as 30 μm. So far, neuroscientists needed to select electrodes manually from hundreds of electrodes. Here we present an electronic depth control algorithm that allows to select electrodes automatically, hereby allowing to reduce the amount of data and locating those electrodes that are close to neurons. The electrodes are selected according to a new penalized signal-to-noise ratio (PSNR) criterion that demotes electrodes from becoming selected if their signals are redundant with previously selected electrodes. It is shown that, using the PSNR, interneurons generating smaller spikes are also selected. We developed a model that aims to evaluate algorithms for electronic depth control, but also generates benchmark data for testing spike sorting and spike detection algorithms. The model comprises a realistic tufted pyramidal cell, non-tufted pyramidal cells and inhibitory interneurons. All neurons are synaptically activated by hundreds of fibers. This arrangement allows the algorithms to be tested in more realistic conditions, including backgrounds of synaptic potentials, varying spike rates with bursting and spike amplitude attenuation.

Comparison of classification methods for P300 brain-computer interface on disabled subjects

We report on tests with a mind typing paradigm based on a P300 brain-computer interface (BCI) on a group of amyotrophic lateral sclerosis (ALS), middle cerebral artery (MCA) stroke, and subarachnoid hemorrhage (SAH) patients, suffering from motor and speech disabilities. We investigate the achieved typing accuracy given the individual patients disorder, and how it correlates with the type of classifier used. We considered 7 types of classifiers, linear as well as nonlinear ones, and found that, overall, one type of linear classifier yielded a higher classification accuracy. In addition to the selection of the classifier, we also suggest and discuss a number of recommendations to be considered when building a P300-based typing system for disabled subjects.

Distinct Mechanisms for Coding of Visual Actions in Macaque Temporal Cortex

Temporal cortical neurons are known to respond to visual dynamic-action displays. Many human psychophysical and functional imaging studies examining biological motion perception have used treadmill walking, in contrast to previous macaque single-cell studies. We assessed the coding of locomotion in rhesus monkey (Macaca mulatta) temporal cortex using movies of stationary walkers, varying both form and motion (i.e., different facing directions) or varying only the frame sequence (i.e., forward vs backward walking). The majority of superior temporal sulcus and inferior temporal neurons were selective for facing direction, whereas a minority distinguished forward from backward walking. Support vector machines using the temporal cortical population responses as input classified facing direction well, but forward and backward walking less so. Classification performance for the latter improved markedly when the within-action response modulation was considered, reflecting differences in momentary body poses within the locomotion sequences. Responses to static pose presentations predicted the responses during the course of the action. Analyses of the responses to walking sequences wherein the start frame was varied across trials showed that some neurons also carried a snapshot sequence signal. Such sequence information was present in neurons that responded to static snapshot presentations and in neurons that required motion. Our data suggest that actions are analyzed by temporal cortical neurons using distinct mechanisms. Most neurons predominantly signal momentary pose. In addition, temporal cortical neurons, including those responding to static pose, are sensitive to pose sequence, which can contribute to the signaling of learned action sequences.

Encoding of Categories by Noncategory-Specific Neurons in the Inferior Temporal Cortex

In order to understand how the brain codes natural categories, e.g., trees and fish, recordings were made in the anterior part of the macaque inferior temporal (IT) cortex while the animal was performing a tree/nontree categorization task. Most single cells responded to exemplars of more than one category while other neurons responded only to a restricted set of exemplars of a given category. Since it is still not known which type of cells contribute and what is the nature of the code used for categorization in IT, we have performed an analysis on single-cell data. A Kohonen self-organizing map (SOM), which uses an unsupervised (competitive) learning algorithm, was used to study the single cell responses to tree and nontree images. Results from the Kohonen SOM indicated that the collected neuronal data consisting of spike counts was sufficient to account for a good level of categorization success (approximately 83) when categorizing a group of 200 trees and nontrees. Contrary to intuition, the results of the investigation suggest that the population of category-specific neurons (neurons that respond only to trees or only to nontrees) was unimportant to the categorization. Instead, a large majority of the neurons that were most important to the categorization was found to belong to a class of more broadly tuned cells, namely, cells that responded to both categories but that favored one category over the other by seven or more images. A simple algebraic operation (without the Kohonen SOM) between the above-mentioned noncategory-specific neurons confirmed the contribution of these neurons to categorization. Thus, the modeling results suggest (1) that broadly tuned neurons are critical for categorization, and (2) that only one additional layer of processing is required to extract the categories from a population of IT neurons.

Joint entropy maximization in kernel-based topographic maps

A new learning algorithm for kernel-based topographic map formation is introduced. The kernel parameters are adjusted individually so as to maximize the joint entropy of the kernel outputs. This is done by maximizing the differential entropies of the individual kernel outputs, given that the maps output redundancy, due to the kernel overlap, needs to be minimized. The latter is achieved by minimizing the mutual information between the kernel outputs. As a kernel, the (radial) incomplete gamma distribution is taken since, for a gaussian input density, the differential entropy of the kernel output will be maximal. Since the theoretically optimal joint entropy performance can be derived for the case of nonoverlapping gaussian mixture densities, a new clustering algorithm is suggested that uses this optimum as its null distribution. Finally, it is shown that the learning algorithm is similar to one that performs stochastic gradient descent on the Kullback-Leibler divergence for a heteroskedastic gaussian mixture density model.

Coding of images of materials by macaque inferior temporal cortical neurons.

Objects vary not only in their shape but also in the material from which they are made. Knowledge of the material properties can contribute to object recognition as well as indicate properties of the object (e.g. ripeness of a fruit). We examined the coding of images of materials by single neurons of the macaque inferior temporal (IT) cortex, an area known to support object recognition and categorization. Stimuli were images of 12 real materials that were illuminated from three different directions. The material textures appeared within five different outline shapes. The majority of responsive IT neurons responded selectively to the material textures, and this selectivity was largely independent of their shape selectivity. The responses of the large majority of neurons were strongly affected by illumination direction. Despite the generally weak illumination‐direction invariance of the responses, Support Vector Machines that used the neural responses as input were able to classify the materials across illumination direction better than by chance. A comparison between the responses to the original images and those to images with a random spectral phase, but matched power spectrum, indicated that the material texture selectivity did not depend merely on differences in the power spectrum but required phase information.