Broad Receptive Fields Research Articles

YOLO-RWY: A Novel Runway Detection Model for Vision-Based Autonomous Landing of Fixed-Wing Unmanned Aerial Vehicles

In scenarios where global navigation satellite systems (GNSSs) and radio navigation systems are denied, vision-based autonomous landing (VAL) for fixed-wing unmanned aerial vehicles (UAVs) becomes essential. Accurate and real-time runway detection in VAL is vital for providing precise positional and orientational guidance. However, existing research faces significant challenges, including insufficient accuracy, inadequate real-time performance, poor robustness, and high susceptibility to disturbances. To address these challenges, this paper introduces a novel single-stage, anchor-free, and decoupled vision-based runway detection framework, referred to as YOLO-RWY. First, an enhanced data augmentation (EDA) module is incorporated to perform various augmentations, enriching image diversity, and introducing perturbations that improve generalization and safety. Second, a large separable kernel attention (LSKA) module is integrated into the backbone structure to provide a lightweight attention mechanism with a broad receptive field, enhancing feature representation. Third, the neck structure is reorganized as a bidirectional feature pyramid network (BiFPN) module with skip connections and attention allocation, enabling efficient multi-scale and across-stage feature fusion. Finally, the regression loss and task-aligned learning (TAL) assigner are optimized using efficient intersection over union (EIoU) to improve localization evaluation, resulting in faster and more accurate convergence. Comprehensive experiments demonstrate that YOLO-RWY achieves AP50:95 scores of 0.760, 0.611, and 0.413 on synthetic, real nominal, and real edge test sets of the landing approach runway detection (LARD) dataset, respectively. Deployment experiments on an edge device show that YOLO-RWY achieves an inference speed of 154.4 FPS under FP32 quantization with an image size of 640. The results indicate that the proposed YOLO-RWY model possesses strong generalization and real-time capabilities, enabling accurate runway detection in complex and challenging visual environments, and providing support for the onboard VAL systems of fixed-wing UAVs.

Wide-field imaging and recognition through cascaded complex scattering media

Considering the obvious application value in the field of minimally invasive and non-destructive clinical healthcare, we explore the challenge of wide-field imaging and recognition through cascaded complex scattering media, a topic that has been less researched, by realizing wide-field imaging and pathological screening through multimode fibers (MMF) and turbid media. To address the challenge of extracting features from chaotic and globally correlated speckles formed by transmitting images through cascaded complex scattering media, we establish a deep learning approach based on SMixerNet. By efficiently using the parameter-free matrix transposition, SMixerNet achieves a broad receptive field with less inductive bias through concise multi-layer perceptron (MLP). This approach circumvents the parameter's intensive requirements of previous implementations relying on self-attention mechanisms for global receptive fields. Imaging and pathological screening results based on extensive datasets demonstrate that our approach achieves better performance with fewer learning parameters, which helps deploy deep learning models on desktop-level edge computing devices for clinical healthcare. Our research shows that, deep learning facilitates imaging and recognition through cascaded complex scattering media. This research extends the scenarios of medical and industrial imaging, offering additional possibilities in minimally invasive and non-destructive clinical healthcare and industrial monitoring in harsh and complex scenarios.

Enhanced noise resilience in passive tone detection via broad-receptive field complex-valued convolutional neural networks.

Tone detection is crucial for passive sonar systems. Numerous algorithms have been developed for passive tone detection, but their effectiveness in detecting weak tones is still limited. To enhance noise resilience in passive tone detection, a broad-receptive field complex-valued structure named attention-driven complex-valued U-Net is proposed. Concretely, two attention mechanisms, namely, temporal attention and harmonic attention, are proposed to broaden the receptive field with high computational efficiency. Complex-valued operators are then introduced to mine both amplitude and phase information of tones. Additionally, a symmetric downsampling and upsampling strategy is proposed to improve the reconstruction accuracy of detailed time-frequency information. Overall, the proposed method demonstrates a strong robustness to noise and a strong ability to generalize. Experimental results on both simulated data and real-world data validate the superiority of the proposed attention-driven complex-valued U-Net against conventional U-shaped structures.

Image Reconstruction for Accelerated MR Scan With Faster Fourier Convolutional Neural Networks.

High quality image reconstruction from undersampled k -space data is key to accelerating MR scanning. Current deep learning methods are limited by the small receptive fields in reconstruction networks, which restrict the exploitation of long-range information, and impede the mitigation of full-image artifacts, particularly in 3D reconstruction tasks. Additionally, the substantial computational demands of 3D reconstruction considerably hinder advancements in related fields. To tackle these challenges, we propose the following: 1) A novel convolution operator named Faster Fourier Convolution (FasterFC), aims at providing an adaptable broad receptive field for spatial domain reconstruction networks with fast computational speed. 2) A split-slice strategy that substantially reduces the computational load of 3D reconstruction, enabling high-resolution, multi-coil, 3D MR image reconstruction while fully utilizing inter-layer and intra-layer information. 3) A single-to-group algorithm that efficiently utilizes scan-specific and data-driven priors to enhance k -space interpolation effects. 4) A multi-stage, multi-coil, 3D fast MRI method, called the faster Fourier convolution based single-to-group network (FAS-Net), comprising a single-to-group k -space interpolation algorithm and a FasterFC-based image domain reconstruction module, significantly minimizes the computational demands of 3D reconstruction through split-slice strategy. Experimental evaluations conducted on the NYU fastMRI and Stanford MRI Data datasets reveal that the FasterFC significantly enhances the quality of both 2D and 3D reconstruction results. Moreover, FAS-Net, characterized as a method that can achieve high-resolution (320, 320, 256), multi-coil, (8 coils), 3D fast MRI, exhibits superior reconstruction performance compared to other state-of-the-art 2D and 3D methods.

Effective Deep Learning‐Based Infrared Spectral Gas Identification Method

AbstractIn order to detect infrared spectral gas components fast and correctly, an improved dilation residual module is proposed in this study by substituting the classic convolution module with the dilation convolution to have a broad receptive field. Based on the residual network, an efficient and effective dilation residual network called DA‐Resnet12 is developed for infrared spectral gas identification by reducing the size of the convolution kernel and the number of dilation convolution modules. The classification accuracy, training duration, and model parametric size are employed as assessment indices. The experimental results reveal that the proposed DA‐ResNet12 network outperforms other comparative methods in terms of model parameter number, accuracy, and time efficiency, proving the efficacy and efficiency of the proposed DA‐ResNet12 network model.

Gain modulation and odor concentration invariance in early olfactory networks.

The broad receptive field of the olfactory receptors constitutes the basis of a combinatorial code that allows animals to detect and discriminate many more odorants than the actual number of receptor types that they express. One drawback is that high odor concentrations recruit lower affinity receptors which can lead to the perception of qualitatively different odors. Here we addressed the contribution that signal-processing in the antennal lobe makes to reduce concentration dependence in odor representation. By means of calcium imaging and pharmacological approach we describe the contribution that GABA receptors play in terms of the amplitude and temporal profiles of the signals that convey odor information from the antennal lobes to higher brain centers. We found that GABA reduces the amplitude of odor elicited signals and the number of glomeruli that are recruited in an odor-concentration-dependent manner. Blocking GABA receptors decreases the correlation among glomerular activity patterns elicited by different concentrations of the same odor. In addition, we built a realistic mathematical model of the antennal lobe that was used to test the viability of the proposed mechanisms and to evaluate the processing properties of the AL network under conditions that cannot be achieved in physiology experiments. Interestingly, even though based on a rather simple topology and cell interactions solely mediated by GABAergic lateral inhibitions, the AL model reproduced key features of the AL response upon different odor concentrations and provides plausible solutions for concentration invariant recognition of odors by artificial sensors.

Skin lesion segmentation using two-phase cross-domain transfer learning framework

Background and ObjectiveDeep learning (DL) models have been used for medical imaging for a long time but they did not achieve their full potential in the past because of insufficient computing power and scarcity of training data. In recent years, we have seen substantial growth in DL networks because of improved technology and an abundance of data. However, previous studies indicate that even a well-trained DL algorithm may struggle to generalize data from multiple sources because of domain shifts. Additionally, ineffectiveness of basic data fusion methods, complexity of segmentation target and low interpretability of current DL models limit their use in clinical decisions. To meet these challenges, we present a new two-phase cross-domain transfer learning system for effective skin lesion segmentation from dermoscopic images. MethodsOur system is based on two significant technical inventions. We examine a two- phase cross-domain transfer learning approach, including model-level and data-level transfer learning, by fine-tuning the system on two datasets, MoleMap and ImageNet. We then present nSknRSUNet, a high-performing DL network, for skin lesion segmentation using broad receptive fields and spatial edge attention feature fusion. We examine the trained model's generalization capabilities on skin lesion segmentation to quantify these two inventions. We cross-examine the model using two skin lesion image datasets, MoleMap and HAM10000, obtained from varied clinical contexts. ResultsAt data-level transfer learning for the HAM10000 dataset, the proposed model obtained 94.63% of DSC and 99.12% accuracy. In cross-examination at data-level transfer learning for the Molemap dataset, the proposed model obtained 93.63% of DSC and 97.01% of accuracy. ConclusionNumerous experiments reveal that our system produces excellent performance and improves upon state-of-the-art methods on both qualitative and quantitative measures.

Droid-MCFG: Android malware detection system using manifest and control flow traces with multi-head temporal convolutional network

Android is the most popular mobile operating system, making it the main target of malware attacks. Machine learning-based attack detection techniques have recently emerged as promising methods that relies heavily on particular features to classify malware. Despite machine learning-based malware detectors having hundreds of features, attackers can use feature-related expertise to generate malware variants to avoid detection. Therefore, the Android security team must constantly develop novel features to detect suspicious attacks. This paper proposes a novel malware detection method called Droid-MCFG that combines the Android features of manifest and Control Flow Graph (CFG). First, reverse engineering tools are used to mine manifest files and Java source codes from Android Package Kit (APK). Second, to represent Android apps with elevated features, we develop a features selection method that retrieves API calls and API sequences from CFGs. The API calls and manifest information are then combined to produce digital fingerprints of Android app actions. Third, a transfer learning approach based on word2vec is developed to extract trained features from digital fingerprints. To thoroughly analyze the novel features, the word2vec is fine-tuned with random, static, and dynamic strategies. Finally, the multi-head Temporal Convolutional Network (TCN) is designed to identify malware based on fine-tuned features. The TCN employs casual convolutions and dilations due to its temporality and broad receptive fields, making it very responsive to API-call sequences and malware activities in the manifest file. The proposed method achieves a classification accuracy of 96.24% using the CICInvesAndMal2019 dataset.

Certain retinal horizontal cells have a center-surround antagonistic organization.

Retinal horizontal cells form a broad receptive field, which contributes to generating antagonistic surround responses in retinal bipolar cells. Here, I report that certain horizontal cells themselves have center-surround antagonistic receptive fields. The receptive fields of yellow/red, blue-type horizontal cells (Y/RB HCs) in the carp retina were measured by the response to the slit of light stimulus using the conventional intracellular electrode. A center stimulus of monochromatic light of 500 nm hyperpolarized Y/RB HCs, whereas the peripheral light depolarized the cells, suggesting that these cells exhibit an antagonistic receptive field at 500 nm light. The length constant of Y/RB HC's depolarizing responses to 600 nm light was 1.22 ± 0.08 mm, which was larger than that (0.61 ± 0.06 mm) of hyperpolarizing responses to 500 nm light. Thus, depolarizing responses of Y/RB HCs exhibit a larger receptive field than hyperpolarizing responses. The length constant of hyperpolarizing responses of luminosity-type HCs (LHCs) was 1.19 ± 0.07 mm, which was similar to that of 500 nm depolarizing responses of Y/RB HCs (1.34 ± 0.11 mm). Depolarizing response of Y/RB HCs was decreased by bath application of GABA and picrotoxin, a GABA receptor antagonist, suggesting that GABAergic signaling may modulate center-surround antagonistic mechanisms in Y/RB HCs. Bipolar cells display center-surround antagonistic receptive fields that play important roles to improve visual contrast. Wide receptive fields of HCs contribute to generating surround responses in bipolar cells. Therefore, the response polarity of Y/RB HCs may affect the width of the surround receptive field in bipolar cells.NEW & NOTEWORTHY Retinal horizontal cells form a broad receptive field, which contributes to generating antagonistic surround responses in retinal bipolar cells. Here, I found that depolarizing responses of yellow/red, blue-type horizontal cells (Y/RB HCs) exhibit a larger receptive field than hyperpolarizing responses at monochromatic lights between 480 nm and 520 nm. Because bipolar cells play a key role in the detection of visual contrast, depolarization or hyperpolarization of Y/RB HCs may regulate the size of the surround receptive field in the bipolar cells.

Transformation of primary sensory cortical representations from layer 4 to layer 2

Sensory input arrives from thalamus in cortical layer (L) 4, which outputs predominantly to superficial layers. L4 to L2 thus constitutes one of the earliest cortical feedforward networks. Despite extensive study, the transformation performed by this network remains poorly understood. We use two-photon calcium imaging to record neural activity in L2-4 of primary vibrissal somatosensory cortex (vS1) as mice perform an object localization task with two whiskers. Touch responses sparsen and become more reliable from L4 to L2, with nearly half of the superficial touch response confined to ~1 % of excitatory neurons. These highly responsive neurons have broad receptive fields and can more accurately decode stimulus features. They participate disproportionately in ensembles, small subnetworks with elevated pairwise correlations. Thus, from L4 to L2, cortex transitions from distributed probabilistic coding to sparse and robust ensemble-based coding, resulting in more efficient and accurate representations.

Review and analysis of COVID-19 lung lesion segmentation technique and algorithms

COVID-19 is a rapidly spreading disease over the world, yet hospital resources are limited. A recent COVID-19 study reveals that CT imaging can be used to detect disease progression and aid diagnosis, as well as to better understand the disease. Many new studies suggest that deep learning could be used to swiftly and effectively detect COVID-19 in chest CT scans. The problem of automatically segmenting the lungs in CT scans of individuals with confirmed or suspected COVID-19 is intriguing. CT chest scan images allow us to analyze and categorize COVID-19's normal and abnormal properties in a comprehensive way. Segmenting and detecting suspicious areas in the images is done by the platform, which then assesses these regions to get an accurate categorization. Deep learning models can benefit from broad receptive fields that can learn contextual information and COVID-19 infection-related features for more accurate segmentation results. Deep learning models. More than 1800 annotated CT slices were used to construct and evaluate LungINFseg. For this purpose, we tested LungINFseg against 13 current deep learning-based segmentation techniques.

Horizontal Cell Feedback to Cone Photoreceptors in Mammalian Retina: Novel Insights From the GABA-pH Hybrid Model.

How neurons in the eye feed signals back to photoreceptors to optimize sensitivity to patterns of light appears to be mediated by one or more unconventional mechanisms. Via these mechanisms, horizontal cells control photoreceptor synaptic gain and enhance key aspects of temporal and spatial center-surround receptive field antagonism. After the transduction of light energy into an electrical signal in photoreceptors, the next key task in visual processing is the transmission of an optimized signal to the follower neurons in the retina. For this to happen, the release of the excitatory neurotransmitter glutamate from photoreceptors is carefully regulated via horizontal cell feedback, which acts as a thermostat to keep the synaptic transmission in an optimal range during changes to light patterns and intensities. Novel findings of a recently described model that casts a classical neurotransmitter system together with ion transport mechanisms to adjust the alkaline milieu outside the synapse are reviewed. This novel inter-neuronal messaging system carries feedback signals using two separate, but interwoven regulated systems. The complex interplay between these two signaling modalities, creating synaptic modulation-at-a-distance, has obscured it’s being defined. The foundations of our understanding of the feedback mechanism from horizontal cells to photoreceptors have been long established: Horizontal cells have broad receptive fields, suitable for providing surround inhibition, their membrane potential, a function of stimulus intensity and size, regulates inhibition of photoreceptor voltage-gated Ca2+ channels, and strong artificial pH buffering eliminates this action. This review compares and contrasts models of how these foundations are linked, focusing on a recent report in mammals that shows tonic horizontal cell release of GABA activating Cl− and HCO3− permeable GABA autoreceptors. The membrane potential of horizontal cells provides the driving force for GABAR-mediated HCO3− efflux, alkalinizing the cleft when horizontal cells are hyperpolarized by light or adding to their depolarization in darkness and contributing to cleft acidification via NHE-mediated H+ efflux. This model challenges interpretations of earlier studies that were considered to rule out a role for GABA in feedback to cones.

Differential Inhibitory Configurations Segregate Frequency Selectivity in the Mouse Inferior Colliculus.

Receptive fields and tuning curves of sensory neurons represent the neural substrates that allow animals to efficiently detect and distinguish external stimuli. They are progressively refined to create diverse sensitivity and selectivity for neurons along ascending central pathways. However, the neural circuitry mechanisms have not been directly determined for such fundamental qualities in relation to sensory neurons' functional organizations, because of the technical difficulty of correlating neurons' input and output. Here, we obtained spike outputs and synaptic inputs from the same neurons within characteristically defined neural ensembles, to determine the synaptic mechanisms driving their diverse frequency selectivity in the mouse inferior colliculus. We find that the synaptic strength and timing of excitatory and inhibitory inputs are configured differently and independently within individual neurons' receptive fields, which segregate sensitive and selective neurons and endow neural populations with broad receptive fields and sharp frequency tuning. By computationally modeling spike outputs from integrating synaptic inputs and comparing them with real spike responses of the same neurons, we show that space-clamping errors did not qualitatively affect the estimation of spike responses derived from synaptic currents in in vivo voltage-clamp recordings. These data suggest that heterogeneous inhibitory circuits coexist locally for a parallel but differentiated representation of incoming signals.SIGNIFICANCE STATEMENT Sensitivity and selectivity are functional qualities of sensory systems to facilitate animals' survival. There is little direct evidence for the synaptic basis of neurons' functional variance within neural ensembles. Here we adopted a novel framework to fill such a long-standing gap by uniting population activities with single cells' spike outputs and their synaptic inputs. Furthermore, the effects of space-clamping errors on subcortical synaptic currents were evaluated in vivo, by comparing recorded spike responses and simulated spike outputs from computationally integrating synaptic inputs. Our study illustrated that the synaptic strength and timing of inhibition relative to excitation can be configured differently for neurons within a defined neural ensemble, to segregate their selectivity. It provides new insights into coexisting heterogeneous local circuits.

Visual-motor response fields and spatial tuning in supplementary eye field (SEF) of the head unrestrained monkeys.

We recently used an analytic technique involving statistical fits of different spatial models against neural response fields to show that visual response encodes target direction and motor response encodes planned gaze direction (both relative to initial eye orientation) in the superior colliculus (SC; Sadeh et al. European journal of Neuroscience.2015) and in the frontal eye fields (FEF; Sajad et al. Cerebral Cortex.2015). Here, we applied a similar methodology to investigate if similar spatial codes are employed in the supplementary eye field (SEF). Since the SEF and FEF have strong reciprocal connections, we hypothesized that they would show similar egocentric spatial codes. Monkeys were trained to make gaze shifts toward briefly presented targets distributed throughout the potential neuronal receptive fields. After a variable memory delay, a go signal was provided for a gaze saccade. Experiments were performed in head-unrestrained conditions with 3-D eye and head recordings to allow discrimination of all egocentric models of gaze coding. Until now, approximately 60 neurons in monkey have been recorded in relation to the above task, 20 with visual responses, 17 with motor responses, and 5 Visuomotor. These have been analyzed employing our previous methodology, including the use of non-parametric fits to response fields. Preliminary data show that both the visual and motor response fields of most SEF neurons show broad receptive fields with poorly defined spatial tuning, and considerable variability in activity for each target direction unrelated to any measured spatial parameter. To date (unlike the FEF), SEF visual population shows a slight (non-significant) bias toward a target within space (Ts) spatial model, but only one model (target relative to head) was statistically eliminated. These preliminary data suggest that the SEF plays a minor role in specifying the spatial direction of targets and gaze plans compared with the SC and FEF. Meeting abstract presented at VSS 2017

Position Information Encoded by Population Activity in Hierarchical Visual Areas.

Neurons in high-level visual areas respond to more complex visual features with broader receptive fields (RFs) compared to those in low-level visual areas. Thus, high-level visual areas are generally considered to carry less information regarding the position of seen objects in the visual field. However, larger RFs may not imply loss of position information at the population level. Here, we evaluated how accurately the position of a seen object could be predicted (decoded) from activity patterns in each of six representative visual areas with different RF sizes [V1–V4, lateral occipital complex (LOC), and fusiform face area (FFA)]. We collected functional magnetic resonance imaging (fMRI) responses while human subjects viewed a ball randomly moving in a two-dimensional field. To estimate population RF sizes of individual fMRI voxels, RF models were fitted for individual voxels in each brain area. The voxels in higher visual areas showed larger estimated RFs than those in lower visual areas. Then, the ball’s position in a separate session was predicted by maximum likelihood estimation using the RF models of individual voxels. We also tested a model-free multivoxel regression (support vector regression, SVR) to predict the position. We found that regardless of the difference in RF size, all visual areas showed similar prediction accuracies, especially on the horizontal dimension. Higher areas showed slightly lower accuracies on the vertical dimension, which appears to be attributed to the narrower spatial distributions of the RF centers. The results suggest that much position information is preserved in population activity through the hierarchical visual pathway regardless of RF sizes and is potentially available in later processing for recognition and behavior.

Differences in the strength of cortical and brainstem inputs to SSA and non-SSA neurons in the inferior colliculus.

In an ever changing auditory scene, change detection is an ongoing task performed by the auditory brain. Neurons in the midbrain and auditory cortex that exhibit stimulus-specific adaptation (SSA) may contribute to this process. Those neurons adapt to frequent sounds while retaining their excitability to rare sounds. Here, we test whether neurons exhibiting SSA and those without are part of the same networks in the inferior colliculus (IC). We recorded the responses to frequent and rare sounds and then marked the sites of these neurons with a retrograde tracer to correlate the source of projections with the physiological response. SSA neurons were confined to the non-lemniscal subdivisions and exhibited broad receptive fields, while the non-SSA were confined to the central nucleus and displayed narrow receptive fields. SSA neurons receive strong inputs from auditory cortical areas and very poor or even absent projections from the brainstem nuclei. On the contrary, the major sources of inputs to the neurons that lacked SSA were from the brainstem nuclei. These findings demonstrate that auditory cortical inputs are biased in favor of IC synaptic domains that are populated by SSA neurons enabling them to compare top-down signals with incoming sensory information from lower areas.

Collagen fibers induce expansion of receptive field of Pacinian corpuscles

Pacinian corpuscles within the skin are mechanoreceptors with a high sensitivity and a broad receptive field, and they contribute to texture evaluation and contact detection. They are distributed in small numbers in the subcutaneous tissue, which is a deep layer in the skin. We propose a new skin model with texture of collagen fibers, based on observations of the thumb of a Macaca fuscata. A finite element analysis using the model reveals that collagen fibers disperse stress concentrations in the subcutaneous tissue, leading to a broad receptive field for Pacinian corpuscles while maintaining their high sensitivity.

Cortical fosGFP Expression Reveals Broad Receptive Field Excitatory Neurons Targeted by POm

Neighboring cortical excitatory neurons show considerable heterogeneity in their responses to sensory stimulation. We hypothesized that a subset of layer 2 excitatory neurons in the juvenile (P18 to 27) mouse whisker somatosensory cortex, distinguished by expression of the activity-dependent fosGFP reporter gene, would be preferentially activated by whisker stimulation. In fact, two-photon targeted, dual whole-cell recordings showed that principal whisker stimulation elicits similar amplitude synaptic responses in fosGFP-expressing and fosGFP(-) neurons. FosGFP(+) neurons instead displayed shorter latency and larger amplitude subthreshold responses to surround whisker stimulation. Using optogenetic stimulation, we determined that these neurons are targeted by axons from the posteromedial nucleus (POm), a paralemniscal thalamic nucleus associated with broad receptive fields and widespread cortical projections. We conclude that fosGFP expression discriminates between single- and multi-whisker receptive field layer 2 pyramidal neurons.

Spatial attention in written word perception

The role of attention in visual word recognition and reading aloud is a long debated issue. Studies of both developmental and acquired reading disorders provide growing evidence that spatial attention is critically involved in word reading, in particular for the phonological decoding of unfamiliar letter strings. However, studies on healthy participants have produced contrasting results. The aim of this study was to investigate how the allocation of spatial attention may influence the perception of letter strings in skilled readers. High frequency words (HFWs), low frequency words and pseudowords were briefly and parafoveally presented either in the left or the right visual field. Attentional allocation was modulated by the presentation of a spatial cue before the target string. Accuracy in reporting the target string was modulated by the spatial cue but this effect varied with the type of string. For unfamiliar strings, processing was facilitated when attention was focused on the string location and hindered when it was diverted from the target. This finding is consistent the assumptions of the CDP+ model of reading aloud, as well as with familiarity sensitivity models that argue for a flexible use of attention according with the specific requirements of the string. Moreover, we found that processing of HFWs was facilitated by an extra-large focus of attention. The latter result is consistent with the hypothesis that a broad distribution of attention is the default mode during reading of familiar words because it might optimally engage the broad receptive fields of the highest detectors in the hierarchical system for visual word recognition.

Self-organized lateral inhibition improves odor classification in an olfaction-inspired network

The insect olfactory system is capable of classifying odorants by encoding and processing the neural representations of chemical stimuli. Odors are transformed into a neuronal representation by a number of receptor classes, each of which encodes a certain combination of chemical features. Those representations resemble a multivariate representation of the stimulus space [1]. The insect olfactory system thus provides an efficient basis for bio-inspired computational methods to process and classify multivariate data. Olfactory receptors typically have broad receptive fields, and the odor spectra of individual receptor classes overlap. From the viewpoint of multivariate data processing, overlapping receptive fields cause correlation between input variables (channel correlation). In previous work, we demonstrated how lateral inhibition in an olfaction-inspired network reduced channel correlation [2,3]. Decorrelation was achieved by setting the strength of lateral inhibition between two channels according to their correlation, which we pre-computed from the input data. Here, we propose unsupervised learning of the lateral inhibition structure. The lateral inhibition synapses support inhibitory spike-timing dependent plasticity (iSTDP) [4,5]. After exposing the network to a sufficient number of input samples, the inhibitory connectivity self-organizes to reflect the correlation between input channels. We show that this biologically realistic, local learning rule produces an inhibitory connectivity that effectively reduces channel correlation and yields superior network performance in a multivariate scent recognition scenario.