Raul C. Mureşan

Pattern recognition using pulse-coupled neural networks and discrete Fourier transforms

Abstract A novel method for pattern recognition using discrete Fourier transforms on the global pulse signal of a pulse-coupled neural network (PCNN) is presented in this paper. We describe the mathematical model of the PCNN and an original way of analyzing the pulse of the network in order to achieve scale- and translation-independent recognition for isolated objects. We also analyze the error as a result of rotation. The system is used for recognizing simple geometric shapes and letters.

Scaled correlation analysis: a better way to compute a cross-correlogram

When computing a cross‐correlation histogram, slower signal components can hinder the detection of faster components, which are often in the research focus. For example, precise neuronal synchronization often co‐occurs with slow co‐variation in neuronal rate responses. Here we present a method – dubbed scaled correlation analysis – that enables the isolation of the cross‐correlation histogram of fast signal components. The method computes correlations only on small temporal scales (i.e. on short segments of signals such as 25 ms), resulting in the removal of correlation components slower than those defined by the scale. Scaled correlation analysis has several advantages over traditional filtering approaches based on computations in the frequency domain. Among its other applications, as we show on data from cat visual cortex, the method can assist the studies of precise neuronal synchronization.

Membrane Resonance Enables Stable and Robust Gamma Oscillations

Neuronal mechanisms underlying beta/gamma oscillations (20–80 Hz) are not completely understood. Here, we show that in vivo beta/gamma oscillations in the cat visual cortex sometimes exhibit remarkably stable frequency even when inputs fluctuate dramatically. Enhanced frequency stability is associated with stronger oscillations measured in individual units and larger power in the local field potential. Simulations of neuronal circuitry demonstrate that membrane properties of inhibitory interneurons strongly determine the characteristics of emergent oscillations. Exploration of networks containing either integrator or resonator inhibitory interneurons revealed that: (i) Resonance, as opposed to integration, promotes robust oscillations with large power and stable frequency via a mechanism called RING (Resonance INduced Gamma); resonance favors synchronization by reducing phase delays between interneurons and imposes bounds on oscillation cycle duration; (ii) Stability of frequency and robustness of the oscillation also depend on the relative timing of excitatory and inhibitory volleys within the oscillation cycle; (iii) RING can reproduce characteristics of both Pyramidal INterneuron Gamma (PING) and INterneuron Gamma (ING), transcending such classifications; (iv) In RING, robust gamma oscillations are promoted by slow but are impaired by fast inputs. Results suggest that interneuronal membrane resonance can be an important ingredient for generation of robust gamma oscillations having stable frequency.

Visual Exploration and Object Recognition by Lattice Deformation

Mechanisms of explicit object recognition are often difficult to investigate and require stimuli with controlled features whose expression can be manipulated in a precise quantitative fashion. Here, we developed a novel method (called “Dots”), for generating visual stimuli, which is based on the progressive deformation of a regular lattice of dots, driven by local contour information from images of objects. By applying progressively larger deformation to the lattice, the latter conveys progressively more information about the target object. Stimuli generated with the presented method enable a precise control of object-related information content while preserving low-level image statistics, globally, and affecting them only little, locally. We show that such stimuli are useful for investigating object recognition under a naturalistic setting – free visual exploration – enabling a clear dissociation between object detection and explicit recognition. Using the introduced stimuli, we show that top-down modulation induced by previous exposure to target objects can greatly influence perceptual decisions, lowering perceptual thresholds not only for object recognition but also for object detection (visual hysteresis). Visual hysteresis is target-specific, its expression and magnitude depending on the identity of individual objects. Relying on the particular features of dot stimuli and on eye-tracking measurements, we further demonstrate that top-down processes guide visual exploration, controlling how visual information is integrated by successive fixations. Prior knowledge about objects can guide saccades/fixations to sample locations that are supposed to be highly informative, even when the actual information is missing from those locations in the stimulus. The duration of individual fixations is modulated by the novelty and difficulty of the stimulus, likely reflecting cognitive demand.

EEG under anesthesia-Feature extraction with TESPAR

We investigated the problem of automatic depth of anesthesia (DOA) estimation from electroencephalogram (EEG) recordings. We employed Time Encoded Signal Processing And Recognition (TESPAR), a time-domain signal processing technique, in combination with multi-layer perceptrons to identify DOA levels. The presented system learns to discriminate between five DOA classes assessed by human experts whose judgements were based on EEG mid-latency auditory evoked potentials (MLAEPs) and clinical observations. We found that our system closely mimicked the behavior of the human expert, thus proving the utility of the method. Further analyses on the features extracted by our technique indicated that information related to DOA is mostly distributed across frequency bands and that the presence of high frequencies (> 80 Hz), which reflect mostly muscle activity, is beneficial for DOA detection.

Properties of multivariate data investigated by fractal dimensionality.

Elaborated data-mining techniques are widely available today. Nevertheless, many non-linear relations among variables remain undiscovered in multi-dimensional datasets. To address this issue we propose a method based on the concept of fractal dimension that explores the structure of multivariate data and apply the method to simulated data, as well as to local field potentials recorded from cat visual cortex. We find that with changes in the analysis scale, the dimensionality of the data often changes, indicating first that the data are not simple fractals with one unique dimension and second, that, at a certain scale, important changes in the geometric structure of the data may occur. The method can be used as a data-mining tool but also as a method for testing a models fit to the data. We achieve the latter by comparing the dimensionality of the original data to the dimensionality of the data reconstructed from a models description of the data (here using the general linear model). The method provides indispensable help in estimating the complexity of non-linear relationships within multivariate datasets.

Visual scale independence in a network of spiking neurons

The scale independence in visual recognition tasks is still one big problem in neurocomputing today. This paper presents a method of obtaining scale independence in a purely feed-forward way, being able to account for ultra-rapid visual categorization. It used a retinotopic architecture of simple spiking neurons with different types of receptive fields, organized in a hierarchical fashion similar to the mammal visual path. Fast shunting inhibition had been implemented using a rank-order coding similar to that described by Thorpe and Gautrais (1998). Scale independence had been achieved by using different sized end-stopping bar detectors and combining them in a scalable way to produce scale independent response over a given domain. This solution does not conflict with the saliency based models and offers a great robustness to clutter.

Phase precession and recession with STDP and Anti-STDP

We show that standard, Hebbian spike-timing dependent plasticity (STDP) induces the precession of the firing phase of neurons in oscillatory networks, while anti-Hebbian STDP induces phase recession. In networks that are subject to oscillatory inhibition, the intensity of excitatory input relative to the inhibitory one determines whether the phase can precess due to STDP or whether the phase is fixed. This phenomenon can give a very simple explanation to the experimentally-observed hippocampal phase precession. Modulation of STDP can lead, through precession and recession, to the synchronization of the firing of a trained neuron to a target phase.

The coherence theory: simple attentional modulation effects

Abstract We present a novel method of achieving attentional modulation effects, based on the spatial coherence of the stimulus. Such modulator effects are known to occur also at low levels in the visual cortical pathway. We use temporal coding rather than rate-based coding. The temporal coding is biologically plausible and also proves to be noise resistant. Synchrony and asynchrony are estimated in an ultra-rapid fashion, the competition between them leading to attentional modulator effects. Our “Coherence Theory” as a new way of understanding neural processing offers the theoretical framework for our findings.

Timescales of multineuronal activity patterns reflect temporal structure of visual stimuli.

The investigation of distributed coding across multiple neurons in the cortex remains to this date a challenge. Our current understanding of collective encoding of information and the relevant timescales is still limited. Most results are restricted to disparate timescales, focused on either very fast, e.g., spike-synchrony, or slow timescales, e.g., firing rate. Here, we investigated systematically multineuronal activity patterns evolving on different timescales, spanning the whole range from spike-synchrony to mean firing rate. Using multi-electrode recordings from cat visual cortex, we show that cortical responses can be described as trajectories in a high-dimensional pattern space. Patterns evolve on a continuum of coexisting timescales that strongly relate to the temporal properties of stimuli. Timescales consistent with the time constants of neuronal membranes and fast synaptic transmission (5–20 ms) play a particularly salient role in encoding a large amount of stimulus-related information. Thus, to faithfully encode the properties of visual stimuli the brain engages multiple neurons into activity patterns evolving on multiple timescales.