Search

Thousands of species use vocal signals to communicate with one another. Vocalizations carry rich information, yet characterizing and analyzing these complex, high-dimensional signals is difficult and prone to human bias. Moreover, animal vocalizations are ethologically relevant stimuli whose representation by auditory neurons is an important subject of research in sensory neuroscience. A method that can efficiently generate naturalistic vocalization waveforms would offer an unlimited supply of stimuli with which to probe neuronal computations. Although unsupervised learning methods allow for the projection of vocalizations into low-dimensional latent spaces learned from the waveforms themselves, and generative modeling allows for the synthesis of novel vocalizations for use in downstream tasks, we are not aware of any model that combines these tasks to synthesize naturalistic vocalizations in the waveform domain for stimulus playback. In this paper, we demonstrate BiWaveGAN: a bidirectional generative adversarial network (GAN) capable of learning a latent representation of ultrasonic vocalizations (USVs) from mice. We show that BiWaveGAN can be used to generate, and interpolate between, realistic vocalization waveforms. We then use these synthesized stimuli along with natural USVs to probe the sensory input space of mouse auditory cortical neurons. We show that stimuli generated from our method evoke neuronal responses as effectively as real vocalizations, and produce receptive fields with the same predictive power. BiWaveGAN is not restricted to mouse USVs but can be used to synthesize naturalistic vocalizations of any animal species and interpolate between vocalizations of the same or different species, which could be useful for probing categorical boundaries in representations of ethologically relevant auditory signals. NEW & NOTEWORTHY: A new type of artificial neural network is presented that can be used to generate animal vocalization waveforms and interpolate between them to create new vocalizations. We find that our synthetic naturalistic stimuli drive auditory cortical neurons in the mouse equally well and produce receptive field features with the same predictive power as those obtained with natural mouse vocalizations, confirming the quality of the stimuli produced by the neural network. [ABSTRACT FROM AUTHOR]

The biological system is not a perfect system, but it is a relatively complete system. It is difficult to realize the lower power consumption and high parallelism that characterize biological systems if lower-level information pathways are ignored. In this paper, we focus on the K, M and P pathways of visual signal processing from the retina to the lateral geniculate nucleus (LGN). We model the visual system at a fine-grained level to ensure efficient information transmission while minimizing energy use. We also implement a circuit-level distributed parallel computing model on FPGAs. The results show that we are able to transfer information with low energy consumption and high parallelism. The Artix-7 family of xc7a200tsbv484-1 FPGAs can reach a maximum frequency of 200 MHz and a maximum parallelism of 600, and a single receptive field model consumes only 0.142 W of power. This can be useful for building assistive vision systems for small and light devices. [ABSTRACT FROM AUTHOR]

PANet is widely used in various object detection tasks due to its powerful feature expression ability. However, PANet's performance in complex scenarios is subpar, with frequent object omission or misidentification. We find that the reason for this phenomenon is that the receptive field of PANet can't cover sufficient feature information, to deal with drastic changes of source object size. In order to solve this problem, this paper adopts dilated convolution technology and applies it to each parallel branch directly following the PANet network. This method can effectively represent the feature information of objects at different scales by integrating the information from small and large receptive fields into a new feature output. We also introduce residual structure to circumvent the network degradation caused by excessive convolutions. By combining the above methods, we build a new module named PANetW (PANet with Wider Receptive Fields). Taking YOLOX-S as the baseline, we comprehensively evaluated the proposed module PANetW on two datasets, VOC2007 and MSCOCO2017. The test results show that our PANetW achieves a high level of mean average precision (AP). On the VOC2007 dataset, the AP of our PANetW improves by 4.9% to 43.0%; on the MS COCO2017 dataset, the AP of PANetW is as high as 44.3%, far exceeding the current mainstream modules. The experimental results fully demonstrate the effectiveness of our module. [ABSTRACT FROM AUTHOR]

In most neuroscience textbooks, the thalamus is presented as a structure that relays sensory signals from visual, auditory, somatosensory, and gustatory receptors to the cerebral cortex. But the function of the thalamic nuclei goes beyond the simple transfer of information. This is especially true for the second-order nuclei, but also applies to first-order nuclei. First order thalamic nuclei receive information from the periphery, like the dorsal lateral geniculate nucleus (dLGN), which receives a direct input from the retina. In contrast, second order thalamic nuclei, like the pulvinar, receive minor or no input from the periphery, with the bulk of their input derived from cortical areas. The dLGN refines the information received from the retina by temporal decorrelation, thereby transmitting the most relevant signals to the visual cortex. The pulvinar is closely linked to virtually all visual cortical areas, and there is growing evidence that it is necessary for normal cortical processing and for aspects of visual cognition. In this article, we will discuss what we know and do not know about these structures and propose some thoughts based on the knowledge gained during the course of our careers. We hope that these thoughts will arouse curiosity about the visual thalamus and its important role, especially for the next generation of neuroscientists.

Zhiyun Zhang, Hao Zheng, Qingquan Yu, Xianghong Jing Institute of Acupuncture and Moxibustion, China Academy of Chinese Medical Sciences, Beijing, People’s Republic of ChinaCorrespondence: Xianghong Jing, Institute of Acupuncture and Moxibustion, China Academy of Chinese Medical Sciences, Beijing, People’s Republic of China, Email jxhtjb@263.netAbstract: The spinal dorsal horn (SDH) transmits sensory information from the periphery to the brain. Wide dynamic range (WDR) neurons within this relay site play a critical role in modulating and integrating peripheral sensory inputs, as well as the process of central sensitization during pathological pain. This group of spinal multi-receptive neurons has attracted considerable attention in pain research due to their capabilities for encoding the location and intensity of nociception. Meanwhile, transmission, processing, and modulation of incoming afferent information in WDR neurons also establish the underlying basis for investigating the integration of acupuncture and pain signals. This review aims to provide a comprehensive examination of the distinctive features of WDR neurons and their involvement in pain. Specifically, we will examine the regulation of diverse supraspinal nuclei on these neurons and analyze their potential in elucidating the mechanisms of acupuncture analgesia.Keywords: WDR neurons, spinal dorsal horn, pain, central sensitization, receptive fields, DNIC, acupuncture analgesia

Small target detection is an important research direction in the field of computer vision, which is widely used in popular fields such as industrial defect detection, satellite remote sensing image detection. However, in printing defects detection, due to the complex defect background and small target, it is difficult to extract multi-scale features, and the extracted features have less available information. Therefore, this paper improves YOLOv7 and proposes an end-to-end printing defects detection algorithm based on context structure perception and multi-scale feature fusion (CM-YOLOv7). CM-YOLOv7 is mainly composed of a Context Structure Awareness Module (CSAM), a Multi-scale Feature Interaction Module (MFIM) and a Feature Refinement Layer Module (FRLM). Firstly, the CSAM uses multiple convolution kernels of different sizes to obtain features of different receptive fields and enhance multi-scale feature extraction. Secondly, for the extracted multi-scale features, the MFIM adaptively fuses the features of adjacent layers to achieve mutual learning between coarse-grained information and fine-grained information and improves the expression ability of small target features lost after convolution. Finally, in order to refine the edge information of the defect target more effectively and enhance the feature expression ability of the image, an FRLM is designed in MFIM. In the experimental part, this paper utilizes the printing defects detection dataset and DOTA-V1.0 dataset in order to better evaluate the effect of the algorithm. The experimental results show that the CM-YOLOv7 model proposed in this paper has a more accurate detection effect on the small target of printing defects. [ABSTRACT FROM AUTHOR]

The biological system is not a perfect system, but it is a relatively complete system. It is difficult to realize the lower power consumption and high parallelism that characterize biological systems if lower-level information pathways are ignored. In this paper, we focus on the K, M and P pathways of visual signal processing from the retina to the lateral geniculate nucleus (LGN). We model the visual system at a fine-grained level to ensure efficient information transmission while minimizing energy use. We also implement a circuit-level distributed parallel computing model on FPGAs. The results show that we are able to transfer information with low energy consumption and high parallelism. The Artix-7 family of xc7a200tsbv484-1 FPGAs can reach a maximum frequency of 200 MHz and a maximum parallelism of 600, and a single receptive field model consumes only 0.142 W of power. This can be useful for building assistive vision systems for small and light devices.

How many thalamic neurons converge onto a cortical cell? This is an important question, because the organization of thalamocortical projections can influence the cortical architecture.1,2 Here, we estimate the degree of thalamocortical convergence in primary visual cortex by taking advantage of the cortical expansion-neurons within a restricted volume in primary visual cortex have overlapping receptive fields driven by a smaller set of inputs from the lateral geniculate nucleus.3-5 Under these conditions, the measurements of cortical receptive fields in a population can be used to infer the receptive fields of their geniculate inputs and the weights of their projections using non-negative matrix factorization.6 The analysis reveals sparse connectivity,7 where a handful (~2-6) of thalamic inputs account for 90% of the total synaptic weight to a cortical neuron. Together with previous findings,8 these results paint a picture consistent with the idea that convergence of a few inputs partly determine the retinotopy and tuning properties of cortical cells.8-13.

Visual neurons respond selectively to features that become increasingly complex from the eyes to the cortex. Retinal neurons prefer flashing spots of light, primary visual cortical (V1) neurons prefer moving bars, and those in higher cortical areas favor complex features like moving textures. Previously, we showed that V1 simple cell tuning can be accounted for by a basic model implementing temporal prediction – representing features that predict future sensory input from past input (Singer et al., 2018). Here, we show that hierarchical application of temporal prediction can capture how tuning properties change across at least two levels of the visual system. This suggests that the brain does not efficiently represent all incoming information; instead, it selectively represents sensory inputs that help in predicting the future. When applied hierarchically, temporal prediction extracts time-varying features that depend on increasingly high-level statistics of the sensory input.

Extracellular recordings were made from 642 units in the primary visual cortex (V1) of a highly visual marsupial, the Tammar wallaby. The receptive field (RF) characteristics of the cells were objectively estimated using the non-linear input model (NIM), and these were correlated with spike shapes. We found that wallaby cortical units had 68% regular spiking (RS), 12% fast spiking (FS), 4% triphasic spiking (TS), 5% compound spiking (CS) and 11% positive spiking (PS). RS waveforms are most often associated with recordings from pyramidal or spiny stellate cell bodies, suggesting that recordings from these cell types dominate in the wallaby cortex. In wallaby, 70-80% of FS and RS cells had orientation selective RFs and had evenly distributed linear and nonlinear RFs. We found that 47% of wallaby PS units were non-orientation selective and they were dominated by linear RFs. Previous studies suggest that the PS units represent recordings from the axon terminals of non-orientation selective cells originating in the lateral geniculate nucleus (LGN). If this is also true in wallaby, as strongly suggested by their low response latencies and bursty spiking properties, the results suggest that significantly more neurons in wallaby LGN are already orientation selective. In wallaby, less than 10% of recorded spikes had triphasic (TS) or sluggish compound spiking (CS) waveforms. These units had a mixture of orientation selective and non-oriented properties, and their cellular origins remain difficult to classify. [ABSTRACT FROM AUTHOR]

Human perception depends upon internal representations of the environment that help to organize the raw information available from the senses by acting as reference patterns. Internal representations are widely characterized using reverse correlation, a method capable of producing unconstrained estimates of the representation itself, all on the basis of simple responses to random stimuli. Despite its advantages, reverse correlation is often infeasible to apply because of its inefficiency—a very large number of stimulus–response trials are required in order to obtain an accurate estimate. Here, we show that an important source of this inefficiency is small, yet nontrivial, correlations that occur by chance between randomly generated stimuli. We demonstrate in simulation that whitening stimuli to remove such correlations before eliciting responses provides greater than 85% improvement in efficiency for a given estimation quality, as well as a two- to fivefold increase in quality for a given sample size. Moreover, unlike conventional approaches, whitening improves the efficiency of reverse correlation without introducing bias into the estimate, or requiring prior knowledge of the target internal representation. Improving the efficiency of reverse correlation with whitening may enable a broader scope of investigations into the individual variability and potential universality of perceptual mechanisms. [ABSTRACT FROM AUTHOR]

In most neuroscience textbooks, the thalamus is presented as a structure that relays sensory signals from visual, auditory, somatosensory, and gustatory receptors to the cerebral cortex. But the function of the thalamic nuclei goes beyond the simple transfer of information. This is especially true for the second-order nuclei, but also applies to first-order nuclei. First order thalamic nuclei receive information from the periphery, like the dorsal lateral geniculate nucleus (dLGN), which receives a direct input from the retina. In contrast, second order thalamic nuclei, like the pulvinar, receive minor or no input from the periphery, with the bulk of their input derived from cortical areas. The dLGN refines the information received from the retina by temporal decorrelation, thereby transmitting the most “relevant” signals to the visual cortex. The pulvinar is closely linked to virtually all visual cortical areas, and there is growing evidence that it is necessary for normal cortical processing and for aspects of visual cognition. In this article, we will discuss what we know and do not know about these structures and propose some thoughts based on the knowledge gained during the course of our careers. We hope that these thoughts will arouse curiosity about the visual thalamus and its important role, especially for the next generation of neuroscientists.

Classification and recognition tasks performed on photonic hardware-based neural networks often require at least one offline computational step, such as in the increasingly popular reservoir computing paradigm. Removing this offline step can significantly improve the response time and energy efficiency of such systems. We present numerical simulations of different algorithms that utilize ultrafast photonic spiking neurons as receptive fields to allow for image recognition without an offline computing step. In particular, we discuss the merits of event, spike-time and rank-order based algorithms adapted to this system. These techniques have the potential to significantly improve the efficiency and effectiveness of optical classification systems, minimizing the number of spiking nodes required for a given task and leveraging the parallelism offered by photonic hardware.

The descending microcircuit from layer 2/3 (L2/3) to layer 5 (L5) is one of the strongest excitatory pathways in the cortex, presumably forming a core component of its feedforward hierarchy. To date, however, no experiments have selectively tested the impact of L2/3 activity on L5 during active sensation. We used optogenetic, cell-type-specific manipulation of L2/3 neurons in the barrel cortex of actively sensing mice (of either sex) to elucidate the significance of this pathway to sensory coding in L5. Contrary to standard models, activating L2/3 predominantly suppressed spontaneous activity in L5, whereas deactivating L2/3 mainly facilitated touch responses in L5. Somatostatin interneurons are likely important to this suppression because their optogenetic deactivation significantly altered the functional impact of L2/3 onto L5. The net effect of L2/3 was to enhance the stimulus selectivity and expand the range of L5 output. These data imply that the core cortical pathway increases the selectivity and expands the range of cortical output through feedforward inhibition.SIGNIFICANCE STATEMENT The primary sensory cortex contains six distinct layers that interact to form the basis of our perception. While rudimentary patterns of connectivity between the layers have been outlined quite extensively in vitro, functional relationships in vivo, particularly during active sensation, remain poorly understood. We used cell-type-specific optogenetics to test the functional relationship between layer 2/3 and layer 5. Surprisingly, we discovered that L2/3 primarily suppresses cortical output from L5. The recruitment of somatostatin-positive interneurons is likely fundamental to this relationship. The net effect of this translaminar suppression is to enhance the selectivity and expand the range of receptive fields, therefore potentially sharpening the perception of space.

Human pose estimation is a special detection task for small object localization. It requires considering not only global structure but local and fine detail due to variable body poses and complex scenes. However, with the sliding window learning mechanism, the convolutional neural network (CNN) can only see the spatial information in a specific size of receptive field in a certain layer. As the network deepens and the receptive field becomes larger, the network gradually focuses on the global spatial information and loses the perception of local features. To help the deep convolutional neural network have the ability to relearn local information for structure analysis in deeper layers, we propose a layer-channel mixed attention mechanism named integrated attention that can be flexibly embedded into a CNN. Multiple features from the previous layers are aggregated to build attention with synchronously observing different ranges of spatial structures. Through our integrated attention, the network can observe the interdependence between local structures across different receptive fields and more clues can be learned to enhance the expressive power of the network for feature learning. The results of extensive experiments show that the integrated attention mechanism is beneficial to human pose estimation. In particular, the integrated attention can help small networks achieve more accurate predictions and even outperforms larger ones with less computation and parameters. Compared with other attention and keypoint refinement modules, our improvement effect is more stable and better. [ABSTRACT FROM AUTHOR]

The issues in the organization of the human retina are analyzed from the point of view of cybernetics. In particular, hypotheses are formed regarding the conditions of the concentric organization of receptive fields, the extraction of informative features, the possibilities of restructuring the receptive fields and their limitations, the transmission of information through ganglion cells with its binding to the location on the retina, as well as regarding the peculiarities of the visual system in different modes. The conditions for the correct operation of on- and off-centers of receptive fields based on bipolars with a cone as the central zone of excitation and the surrounding rods as an inhibition zone, organized with the help of horizontal cells for the on-center, and vice versa for the off-center, are determined. The conditions for the extraction of informative features based on on- and off-centers in the form of point values of brightness differences between neighboring elements are substantiated. The rigid organization of the receptive fields of ganglion cells and the limited possibilities of its reconstruction are substantiated. A hypothesis on the organization of the transmission of information about brightness differences from bipolar cells through ganglion cells under the control of amacrine cells with reference to the location of the retina is proposed. A hypothesis on the continuous perception of the image by the retina receptors and the discrete perception of information by the brain is formulated. The features of the visual system in the modes of contemplation, searching for an object in an image, tracking it, and detailed examination and recognition are considered. [ABSTRACT FROM AUTHOR]

Computer-aided diagnosis is becoming increasingly important in modern medicine, and computer-aided diagnosis systems require accurate and automatic segmentation of medical images. Due to the unique jump connection structure, U-based convolutional neural networks (CNN) have been widely used in medical image segmentation tasks. However, the CNN-based method has the small receptive field and is not effective in learning relationships between distant features. To address this problem, a depthwise separable convolution network with multiscale large kernel for medical image segmentation (DSML-UNet) is proposed. The depthwise separable property of large kernel depthwise separable convolution increases the receptive field while adding little computational complexity. In the DSML-UNet, different sizes of large kernel depth-separable convolution modules are proposed. The different sizes of large kernel convolution provide multi-scale information and large receptive fields, which can effectively learn the rich semantic information and relationship between distant features. The proposed DSML-UNet is used in the spine dataset (SpineSagT2Wdataset3), the skin dataset (ISIC) and the lung dataset (FML) with Dice Scores of 0.8273, 0.8159, and 0.9581 respectively. The experimental results show that DSML-UNet improves the segmentation performance compared with other related advanced work. [ABSTRACT FROM AUTHOR]

Exploiting data invariances is crucial for efficient learning in both artificial and biological neural circuits. Understanding how neural networks can discover appropriate representations capable of harnessing the underlying symmetries of their inputs is thus crucial in machine learning and neuroscience. Convolutional neural networks, for example, were designed to exploit translation symmetry, and their capabilities triggered the first wave of deep learning successes. However, learning convolutions directly from translation-invariant data with a fully connected network has so far proven elusive. Here we show how initially fully connected neural networks solving a discrimination task can learn a convolutional structure directly from their inputs, resulting in localized, space-tiling receptive fields. These receptive fields match the filters of a convolutional network trained on the same task. By carefully designing data models for the visual scene, we show that the emergence of this pattern is triggered by the non-Gaussian, higher-order local structure of the inputs, which has long been recognized as the hallmark of natural images. We provide an analytical and numerical characterization of the pattern formation mechanism responsible for this phenomenon in a simple model and find an unexpected link between receptive field formation and tensor decomposition of higher-order input correlations. These results provide a perspective on the development of low-level feature detectors in various sensory modalities and pave the way for studying the impact of higher-order statistics on learning in neural networks. [ABSTRACT FROM AUTHOR]

The paper is devoted to issues of perceptual quality of images after compression/recovery processing. In contrast to the known approaches, our investigation has a fundamentally different background. We propose the perceptual coding synthesis based on image representations most adequate for human perception, but not on the most appropriate perceptual metrics. Namely, we base our approach on the previously developed biologically motivated representation of images by samples of counts (sampling representations). The paper gives a short overview of this concept. Since the sampling representations are essentially random, the approach proposed differs from the others in its fundamentally statistical orientation. In this respect, it is very close to the generative model-based approaches currently popular in machine learning. The basic scheme of the approach proposed is focused on generative models in the form of sampling representations, discussed in detail in the paper. Since within the framework of generative models the main instruments are the probability distributions of input/output data, the focus of our approach is also on the probability distributions of counts of sampling representations. As the specific feature of sampling representations is the fact that their complete statistical description has the form of a product of the distribution densities of individual counts, the goal of the approach proposed is, in essence, to estimate these densities. Within the scheme of the approach proposed such estimates concern the parameters of the densities chosen from a certain parametric family of probability distributions. In this regard, the choice of the model of the parametric family that is adequate to the visual perception is extremely important. So, the paper discusses in detail the choice of a parametric family in the form of a system of receptive fields. The theoretical aspects of the approach proposed are supplemented by the results of computer simulation. In conclusion, a summary of the paper discussion is given. [ABSTRACT FROM AUTHOR]

Automatic multimodal image segmentation is considered a challenging research area in the biomedical field. U-shaped models have led to an enormous breakthrough in a large domain of medical image segmentation in recentyears. The receptive field plays an essential role in convolutionalneural networks because too small a receptive field limits context information, and too large loses localization accuracy. Despite outstanding overall performance in biomedical segmenting, classical UNet architecture uses a fixed receptive field in convolutions operations. This study proposes a few modifications in classical UNet architecture by adjusting the receptive field via feature-fused module and attention gate mechanism. Compared with baseline UNet, the numerical parameters of FF-UNet (3.94 million) is 51% of classical UNet architecture (7.75 million). Furthermore, we extended our model performance by introducing post-processing schemes. The tri-threshold fuzzy intensification-based contrast enhancement technique is utilized to improve the contrast of biomedical datasets. In the second tier, the black top-hat filtering-based method is employed to remove hair-like artifacts from the ISIC 2018 skin lesion dataset, which may create a barrier to correctly segmenting the images. The proposed models have been trained using fivefold cross-validation on five publicly available biomedical datasets and achieved the dice coefficients of 0.860, 0.932, 0.932, 0.925, and 0.894 on ETIS-LaribPolypDB, CVC-ColonDB, CVC-ClinicDB, DSB 2018, and ISIC 2018 datasets, respectively. To further verify our claims, comparative analysis based on dice results is conducted, proving the proposed model effectiveness. The FF-UNet implementation models and pre-trained weights are freely publicly available: https://github.com/ahmedeqbal/FF-UNet. [ABSTRACT FROM AUTHOR]

To understand the experience of seeing is difficult and is being pursued day and night, especially in psychology and neuroscience. However, there are hard areas to research, such as estimating the structure of receptive fields (RF) in the higher-order visual cortex with humans and animals as research targets. Deep neural networks (DNNs) are being reported that have similar properties to visual neurons and the possibility of using DNNs as alternative re- search targets to the biological brain has emerged. Therefore, in this paper, I discuss whether DNNs can be our re- search subject. In this research, I applied the reverse correlation method, which has revealed RF of visual neurons, to DNNs to estimate RF of units in well-trained VGG-16. As a result, the properties of the RF of VGG-16 units were similar to visual cortex neurons. The result suggested that DNNs may be a good alternative model for our research, but also suggested limitations of the method. To solve remaining problems, psychology research that develop the better methods and deep learning research that provides better alternative models must go hand in hand. [ABSTRACT FROM AUTHOR]

Real-time medical image segmentation can not only enhance the interactivity and feasibility of applications but also support more medical application scenarios. Local feature extraction methods reliant on Convolutional Neural Networks (CNN) are hampered by restricted receptive fields, which weakens their ability to capture comprehensive information. Conversely, global feature extraction methods based on Transformers generally face impediments in real-time tasks due to their extensive computational demands. To address these challenges and explore accurate and real-time medical image segmentation models, we introduce this novel MFCPNet. MFCPNet begins by devising Multi-Scale Multi-Channel Convolution (MSMC Conv) to extract local features across various levels and scales. This innovative design contributes to extracting richer local information without unduly burdening the model. Second, for the enhanced receptive field of convolution and the model's generalization capability, we introduce an Attention Block (Attn Block) carrying rotation invariance. This block, inspired by lightweight Bi-Level Routing Attention (BRA) and MLP-Mixer, effectively mitigates the constraints of convolutional structures and achieves superior contextual modeling. Finally, a judicious pruning of the channel count is employed within MFCPNet, striking a trade-off between segmentation accuracy and efficiency. To evaluate the proposed method, we compare it with several classic approaches using three different types of datasets: retinal images, brain scans, and colon polyps. Across these datasets, MFCPNet achieves segmentation performance comparable to existing methods, with a computational cost of 2.2G FLOPs and 0.49M parameters. Furthermore, it demonstrates a processing speed of 79.54 FPS, meeting the requirements for real-time applications. • MFCPNet has local and global feature extraction capabilities and real-time speed. • The lightweight MSMC Conv overcomes the limitations of receptive fields to captures multi-scale local features. • The lightweight Attn Block captures global features. • The Coarse-grained pruning method allows the model to reach real-time speed. [ABSTRACT FROM AUTHOR]

Selective visual attention enables organisms to enhance the representation of behaviorally relevant stimuli by altering the encoding properties of single receptive fields (RFs). Yet we know little about how the attentional modulations of single RFs contribute to the encoding of an entire visual scene. Addressing this issue requires (1) measuring a group of RFs that tile a continuous portion of visual space, (2) constructing a population-level measurement of spatial representations based on these RFs, and (3) linking how different types of RF attentional modulations change the population-level representation. To accomplish these aims, we used fMRI to characterize the responses of thousands of voxels in retinotopically organized human cortex. First, we found that the response modulations of voxel RFs (vRFs) depend on the spatial relationship between the RF center and the visual location of the attended target. Second, we used two analyses to assess the spatial encoding quality of a population of voxels. We found that attention increased fine spatial discriminability and representational fidelity near the attended target. Third, we linked these findings by manipulating the observed vRF attentional modulations and recomputing our measures of the fidelity of population codes. Surprisingly, we discovered that attentional enhancements of population-level representations largely depend on position shifts of vRFs, rather than changes in size or gain. Our data suggest that position shifts of single RFs are a principal mechanism by which attention enhances population-level representations in visual cortex.SIGNIFICANCE STATEMENT Although changes in the gain and size of RFs have dominated our view of how attention modulates visual information codes, such hypotheses have largely relied on the extrapolation of single-cell responses to population responses. Here we use fMRI to relate changes in single voxel receptive fields (vRFs) to changes in population-level representations. We find that vRF position shifts contribute more to population-level enhancements of visual information than changes in vRF size or gain. This finding suggests that position shifts are a principal mechanism by which spatial attention enhances population codes for relevant visual information. This poses challenges for labeled line theories of information processing, suggesting that downstream regions likely rely on distributed inputs rather than single neuron-to-neuron mappings.

The signal transformations that take place in high-level sensory regions of the brain remain enigmatic because of the many nonlinear transformations that separate responses of these neurons from the input stimuli. One would like to have dimensionality reduction methods that can describe responses of such neurons in terms of operations on a large but still manageable set of relevant input features. A number of methods have been developed for this purpose, but often these methods rely on the expansion of the input space to capture as many relevant stimulus components as statistically possible. This expansion leads to a lower effective sampling thereby reducing the accuracy of the estimated components. Alternatively, so-called low-rank methods explicitly search for a small number of components in the hope of achieving higher estimation accuracy. Even with these methods, however, noise in the neural responses can force the models to estimate more components than necessary, again reducing the methods' accuracy. Here we describe how a flexible regularization procedure, together with an explicit rank constraint, can strongly improve the estimation accuracy compared to previous methods suitable for characterizing neural responses to natural stimuli. Applying the proposed low-rank method to responses of auditory neurons in the songbird brain, we find multiple relevant components making up the receptive field for each neuron and characterize their computations in terms of logical OR and AND computations. The results highlight potential differences in how invariances are constructed in visual and auditory systems.

Summary: In recent years, finger vein recognition has attracted more attention and research as a secure method of identification. Convolutional neural networks have achieved great success in the field of finger vein recognition, yet they suffer from high computational complexity, large parameters, and other challenges. To solve these problems, we propose a Gabor convolutional neural network with receptive fields. We use Gabor filters with receptive field properties to design Gabor convolutional layers. Then we replace the conventional convolutional layer with the Gabor convolutional layer; analyze the influence of different loss functions, convolution kernel size, and feature size on the network model; and choose the most suitable model parameters and loss function. Finally, we systematically investigate comparative performance using AGCNN and CNNs in different finger vein databases. Experimental results show that the parameter complexity of AGCNN is significantly less than that of CNNs with a slight performance decrease. [ABSTRACT FROM AUTHOR]

Complex resource usage patterns of scaling Cloud workloads and heterogeneous infrastructure remain a challenge for accurate modelling of server load, which is the key to effective capacity sizing and provisioning in data centers. Recently, Long Short-Term Memory (LSTM) network has been used for host load prediction. However, learning complex noisy variations in host load is still an issue that needs to be addressed. In this work, we propose pCNN-LSTM, a hybrid prediction approach comprising of 1-dimensional Convolution Neural Networks (1D CNN) and LSTM, to predict CPU utilization on Cloud servers at multiple consecutive time-steps. It consists of three parallel dilated 1D CNN layers with different dilation rates for pattern extraction from noisy host CPU usage and an LSTM layer that learns temporal dependencies within the raw usage values as well as within the patterns extracted by the 1D CNN layers. Convolutions with different dilation rates enable the model to learn CPU load variations at different scales. Prediction skill of pCNN-LSTM is demonstrated using Google cluster trace, Alibaba trace and Bitbrains data, and performance is measured using Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). pCNN-LSTM achieves up to 15%, 13% and 16% improvements in host load prediction with Google Trace, Alibaba trace and Bitbrains data set, respectively, over LSTM, Bidirectional LSTM (BLSTM), CNN-LSTM, CNN-BLSTM and two of its variants, showing the effectiveness of multi-scale learning capability of pCNN-LSTM and establishes its applicability as an adaptive prediction method for improved capacity planning and provisioning. [ABSTRACT FROM AUTHOR]

Movable tactile sensors in the form of whiskers are present in most mammals, but sensory coding in the cortical whisker representation has been studied almost exclusively in mice and rats. Many species that possess whiskers lack the modular "barrel" organization found in the primary somatosensory cortex (S1) of mice and rats, but it is unclear how whisker-related input is represented in these species. We used single-unit extracellular recording techniques to characterize receptive fields and response properties in S1 of Monodelphis domestica (short-tailed opossum), a nocturnal, terrestrial marsupial that shared its last common ancestor with placental mammals over 160 million years ago. Short-tailed opossums lack barrels and septa in S1 but show active whisking behavior similar to that of mice and rats. Most neurons in short-tailed opossum S1 exhibited multiwhisker receptive fields, including a single best whisker (BW) and lower magnitude responses to the deflection of surrounding whiskers. Mean tuning width was similar to that reported for mice and rats. Both symmetrical and asymmetrical receptive fields were present. Neurons tuned to ventral whiskers tended to show broad tuning along the rostrocaudal axis. Thus, despite the absence of barrels, most receptive field properties were similar to those reported for mice and rats. However, unlike those species, S1 neuronal responses to BW and surround whisker deflection showed comparable latencies in short-tailed opossums. This dissimilarity suggests that some aspects of barrel cortex function may not generalize to tactile processing across mammalian species and may be related to differences in the architecture of the whisker-to-cortex pathway. J. Comp. Neurol. 524:3587-3613, 2016. © 2016 Wiley Periodicals, Inc.

In this article, we consider the problem of reconstructing an image that is downsampled in the space of its SE (2) wavelet transform, which is motivated by classical models of simple cell receptive fields and feature preference maps in the primary visual cortex. We prove that, whenever the problem is solvable, the reconstruction can be obtained by an elementary project and replace iterative scheme based on the reproducing kernel arising from the group structure, and show numerical results on real images. [ABSTRACT FROM AUTHOR]

Perception of quantities, such as numerosity, timing, and size, is essential for behavior and cognition. Accumulating evidence demonstrates neurons processing quantities are tuned, that is, have a preferred quantity amount, not only for numerosity, but also other quantity dimensions and sensory modalities. We argue that quantity-tuned neurons are fundamental to understanding quantity perception. We illustrate how the properties of quantity-tuned neurons can underlie a range of perceptual phenomena. Furthermore, quantity-tuned neurons are organized in distinct but overlapping topographic maps. We suggest that this overlap in tuning provides the neural basis for perceptual interactions between different quantities, without the need for a common neural representational code. Humans and animals share a sense of quantities, such as numerosity, timing, and size, which underlies various cognitive functions. Neural measurements reveal neurons with selective, tuned responses to specific quantities, organized in topographic maps. The tuning properties of quantity-selective neurons link brain and behavior, both explaining and unifying distinct behavioral effects in quantity perception. Traditionally discrete perceptual processes observed in subitizing and estimation numerosity ranges are united into a common framework based on similar neural tuning and neural adaptation effects within the same topographic maps. Perceptual interactions between quantities may emerge from interactions between nearby neural populations, tuned to different quantities and modalities, without mapping these onto a single neural representation. [ABSTRACT FROM AUTHOR]

In this article, we consider the problem of reconstructing an image that is downsampled in the space of its SE(2) wavelet transform, which is motivated by classical models of simple cell receptive fields and feature preference maps in the primary visual cortex. We prove that, whenever the problem is solvable, the reconstruction can be obtained by an elementary project and replace iterative scheme based on the reproducing kernel arising from the group structure, and show numerical results on real images.

Animals have evolved sophisticated visual circuits to solve a vital inference problem: detecting whether or not a visual signal corresponds to an object on a collision course. Such events are detected by specific circuits sensitive to visual looming, or objects increasing in size. Various computational models have been developed for these circuits, but how the collision-detection inference problem itself shapes the computational structures of these circuits remains unknown. Here, inspired by the distinctive structures of LPLC2 neurons in the visual system of Drosophila, we build anatomically-constrained shallow neural network models and train them to identify visual signals that correspond to impending collisions. Surprisingly, the optimization arrives at two distinct, opposing solutions, only one of which matches the actual dendritic weighting of LPLC2 neurons. Both solutions can solve the inference problem with high accuracy when the population size is large enough. The LPLC2-like solutions reproduces experimentally observed LPLC2 neuron responses for many stimuli, and reproduces canonical tuning of loom sensitive neurons, even though the models are never trained on neural data. Thus, LPLC2 neuron properties and tuning are predicted by optimizing an anatomically-constrained neural network to detect impending collisions. More generally, these results illustrate how optimizing inference tasks that are important for an animal’s perceptual goals can reveal and explain computational properties of specific sensory neurons.

Cortical processing of auditory information can be affected by interspecies differences as well as brain states. Here we compare multifeature spectro-temporal receptive fields (STRFs) and associated input/output functions or nonlinearities (NLs) of neurons in primary auditory cortex (AC) of four mammalian species. Single-unit recordings were performed in awake animals (female squirrel monkeys, female, and male mice) and anesthetized animals (female squirrel monkeys, rats, and cats). Neuronal responses were modeled as consisting of two STRFs and their associated NLs. The NLs for the STRF with the highest information content show a broad distribution between linear and quadratic forms. In awake animals, we find a higher percentage of quadratic-like NLs as opposed to more linear NLs in anesthetized animals. Moderate sex differences of the shape of NLs were observed between male and female unanesthetized mice. This indicates that the core AC possesses a rich variety of potential computations, particularly in awake animals, suggesting that multiple computational algorithms are at play to enable the auditory system's robust recognition of auditory events., (© The Author(s) 2024. Published by Oxford University Press.)

Objective : Neurons in primary visual cortex (V1) display a range of sensitivity in their response to translations of their preferred visual features within their receptive field: from high specificity to a precise position through to complete invariance. This visual feature selectivity and invariance is frequently modeled by applying a selection of linear spatial filters to the input image, that define the feature selectivity, followed by a nonlinear function that combines the filter outputs, that defines the invariance, to predict the neural response. We compare two such classes of model, that are both popular and parsimonious, the generalized quadratic model (GQM) and the nonlinear input model (NIM). These two classes of model differ primarily in that the NIM can accommodate a greater diversity in the form of nonlinearity that is applied to the outputs of the filters. Approach : We compare the two model types by applying them to data from multielectrode recordings from cat primary visual cortex in response to spatially white Gaussian noise After fitting both classes of model to a database of 342 single units (SUs), we analyze the qualitative and quantitative differences in the visual feature processing performed by the two models and their ability to predict neural response. Main results : We find that the NIM predicts response rates on a held-out data at least as well as the GQM for 95% of SUs. Superior performance occurs predominantly for those units with above average spike rates and is largely due to the NIMs ability to capture aspects of the model's nonlinear function cannot be captured with the GQM rather than differences in the visual features being processed by the two different models. Significance : These results can help guide model choice for data-driven receptive field modelling., (Creative Commons Attribution license.)

Tactile nerve fibers fall into a few classes that can be readily distinguished based on their spatiotemporal response properties. Because nerve fibers reflect local skin deformations, they individually carry ambiguous signals about object features. In contrast, cortical neurons exhibit heterogeneous response properties that reflect computations applied to convergent input from multiple classes of afferents, which confer to them a selectivity for behaviorally relevant features of objects. The conventional view is that these complex response properties arise within the cortex itself, implying that sensory signals are not processed to any significant extent in the two intervening structures--the cuneate nucleus (CN) and the thalamus. To test this hypothesis, we recorded the responses evoked in the CN to a battery of stimuli that have been extensively used to characterize tactile coding in both the periphery and cortex, including skin indentations, vibrations, random dot patterns, and scanned edges. We found that CN responses are more similar to their cortical counterparts than they are to their inputs: CN neurons receive input from multiple classes of nerve fibers, they have spatially complex receptive fields, and they exhibit selectivity for object features. Contrary to consensus, then, the CN plays a key role in processing tactile information. [ABSTRACT FROM AUTHOR]

Current models of retinogeniculate development have proposed that connectivity between the retina and the dorsal lateral geniculate nucleus (dLGN) is established by gradients of axon guidance molecules, to allow initial coarse connections, and by competitive Hebbian-like processes, to drive eye-specific segregation and refine retinotopy. Here we show that when intereye competition is eliminated by monocular enucleation, blocking cholinergic stage II retinal waves disrupts the intraeye competition-mediated expansion of the retinogeniculate projection and results in the permanent disorganization of its laminae. This disruption of stage II retinal waves also causes long-term impacts on receptive field size and fine-scale retinotopy in the dLGN. Our results reveal a novel role for stage II retinal waves in regulating retinogeniculate afferent terminal targeting by way of intraeye competition, allowing for correct laminar patterning and the even allocation of synaptic territory. These findings should contribute to answering questions regarding the role of neural activity in guiding the establishment of neural circuits.