Feedback Connections Research Articles

Low-spatial-frequency priming potentiates the high-level mechanisms of contextual influence

The context sensitivity of visual size perception can be enhanced by prior exposure to low-spatial-frequency (LSF) relative to high-spatial-frequency information. Whether LSF priming affects low- or high-level mechanisms of contextual influence remains to be clarified. By using the Ebbinghaus illusion, which is a classic example of context-dependent size perception, we reduced the lightness contrast of surrounding inducers relative to the background and the shape similarity of the central target and surrounding inducers to diminish or eliminate low-level contour interaction and high-level size contrast, respectively. The results showed that LSF-related enhancement of context influence was unaffected by a change to the surround elements’ contrast but not by a change to the shape similarity between the target and surround. The findings provide evidence that LSF priming primarily affects high-level rather than low-level mechanisms of contextual influence, possibly by weakening the suppressive function of feedback connections from higher visual regions to the early visual cortex.

Neural Oscillatory Abnormalities During Gaze Processing in Schizophrenia: Evidence of Reduced Theta Phase Consistency and Inter-areal Theta-Gamma Coupling

BackgroundAbnormal gaze discrimination in schizophrenia (SZ) is associated with impairment in social functioning, but the neural mechanisms remain unclear. Evidence suggests that local neural oscillations and inter-areal communication through neural synchronization are critical physiological mechanisms supporting basic and complex cognitive processes. The roles of these mechanisms in abnormal gaze processing in SZ have not been investigated. The present study examined local neural oscillations and connectivity between anterior and bilateral posterior brain areas during gaze processing. MethodsDuring electroencephalography recording, 28 participants with SZ and 34 healthy control participants completed a gaze discrimination task. Time-frequency decomposition of electroencephalography data was used to examine neural oscillatory power and intertrial phase consistency at bilateral posterior and midline anterior scalp sites. In addition, connectivity between these anterior and posterior sites, in terms of cross-frequency coupling between theta phase and gamma amplitude, was examined using the Kullback-Leibler Modulation Index. ResultsParticipants with SZ showed reduced total power of theta-band activity relative to healthy control participants at all sites examined. This group difference could be accounted for by reduced intertrial phase consistency of theta activity in SZ participants, which was related to reduced gaze discrimination accuracy in SZ. In addition, SZ participants exhibited reduced Kullback-Leibler indexing, both feedforward and feedback connectivity, between the posterior and anterior sites. ConclusionsThese findings suggest that abnormal theta phase consistency and dysconnection between posterior face processing and anterior areas may underlie gaze processing deficits in SZ.

Ambulation in Dogs With Absent Pain Perception After Acute Thoracolumbar Spinal Cord Injury

Acute thoracolumbar spinal cord injury (SCI) is common in dogs frequently secondary to intervertebral disc herniation. Following severe injury, some dogs never regain sensory function to the pelvic limbs or tail and are designated chronically “deep pain negative.” Despite this, a subset of these dogs develop spontaneous motor recovery over time including some that recover sufficient function in their pelvic limbs to walk independently without assistance or weight support. This type of ambulation is commonly known as “spinal walking” and can take up to a year or more to develop. This review provides a comparative overview of locomotion and explores the physiology of locomotor recovery after severe SCI in dogs. We discuss the mechanisms by which post-injury plasticity and coordination between circuitry contained within the spinal cord, peripheral sensory feedback, and residual or recovered supraspinal connections might combine to underpin spinal walking. The clinical characteristics of spinal walking are outlined including what is known about the role of patient or injury features such as lesion location, timeframe post-injury, body size, and spasticity. The relationship between the emergence of spinal walking and electrodiagnostic and magnetic resonance imaging findings are also discussed. Finally, we review possible ways to predict or facilitate recovery of walking in chronically deep pain negative dogs. Improved understanding of the mechanisms of gait generation and plasticity of the surviving tissue after injury might pave the way for further treatment options and enhanced outcomes in severely injured dogs.

Layer-dependent multiplicative effects of spatial attention on contrast responses in human early visual cortex.

Attention mechanisms at different cortical layers of human visual cortex remain poorly understood. Using submillimeter-resolution fMRI at 7 Tesla, we investigated the effects of top-down spatial attention on the contrast responses across different cortical depths in human early visual cortex. Gradient echo (GE) T2* weighted BOLD signal showed an additive effect of attention on contrast responses across cortical depths. Compared to the middle cortical depth, attention modulation was stronger in the superficial and deep depths of V1, and also stronger in the superficial depth of V2 and V3. Using ultra-high resolution (0.3 mm in-plane) balanced steady-state free precession (bSSFP) fMRI, a multiplicative scaling effect of attention was found in the superficial and deep layers, but not in the middle layer of V1. Attention modulation of low contrast response was strongest in the middle cortical depths, indicating baseline enhancement or contrast gain of attention modulation on feedforward input. Finally, the additive vs. scaling effect of attention from GE-EPI and bSSFP signals can be explained by stronger nonlinearity of BOLD signals from large than small blood vessels, suggesting multiplicative effect of attention on neural activity. These findings support that top-down spatial attention mainly operates through feedback connections from higher order cortical areas, and a distinct mechanism of attention may also be associated with feedforward input through subcortical pathway.

A Predictive-Coding Network That Is Both Discriminative and Generative.

Predictive coding (PC) networks are a biologically interesting class of neural networks. Their layered hierarchy mimics the reciprocal connectivity pattern observed in the mammalian cortex, and they can be trained using local learning rules that approximate backpropagation (Bogacz, 2017). However, despite having feedback connections that enable information to flow down the network hierarchy, discriminative PC networks are not typically generative. Clamping the output class and running the network to equilibrium yields an input sample that usually does not resemble the training input. This letter studies this phenomenon and proposes a simple solution that promotes the generation of input samples that resemble the training inputs. Simple decay, a technique already in wide use in neural networks, pushes the PC network toward a unique minimum two-norm solution, and that unique solution provably (for linear networks) matches the training inputs. The method also vastly improves the samples generated for nonlinear networks, as we demonstrate on MNIST.

Brain-inspired replay for continual learning with artificial neural networks

Artificial neural networks suffer from catastrophic forgetting. Unlike humans, when these networks are trained on something new, they rapidly forget what was learned before. In the brain, a mechanism thought to be important for protecting memories is the reactivation of neuronal activity patterns representing those memories. In artificial neural networks, such memory replay can be implemented as ‘generative replay’, which can successfully – and surprisingly efficiently – prevent catastrophic forgetting on toy examples even in a class-incremental learning scenario. However, scaling up generative replay to complicated problems with many tasks or complex inputs is challenging. We propose a new, brain-inspired variant of replay in which internal or hidden representations are replayed that are generated by the network’s own, context-modulated feedback connections. Our method achieves state-of-the-art performance on challenging continual learning benchmarks (e.g., class-incremental learning on CIFAR-100) without storing data, and it provides a novel model for replay in the brain.

Does Delta Connectivity Differentiate Sleep and Anesthesia?

Does Delta Connectivity Differentiate Sleep and Anesthesia?

Facilitating the propagation of spiking activity in feedforward networks by including feedback.

Transient oscillations in network activity upon sensory stimulation have been reported in different sensory areas of the brain. These evoked oscillations are the generic response of networks of excitatory and inhibitory neurons (EI-networks) to a transient external input. Recently, it has been shown that this resonance property of EI-networks can be exploited for communication in modular neuronal networks by enabling the transmission of sequences of synchronous spike volleys ('pulse packets'), despite the sparse and weak connectivity between the modules. The condition for successful transmission is that the pulse packet (PP) intervals match the period of the modules' resonance frequency. Hence, the mechanism was termed communication through resonance (CTR). This mechanism has three severe constraints, though. First, it needs periodic trains of PPs, whereas single PPs fail to propagate. Second, the inter-PP interval needs to match the network resonance. Third, transmission is very slow, because in each module, the network resonance needs to build up over multiple oscillation cycles. Here, we show that, by adding appropriate feedback connections to the network, the CTR mechanism can be improved and the aforementioned constraints relaxed. Specifically, we show that adding feedback connections between two upstream modules, called the resonance pair, in an otherwise feedforward modular network can support successful propagation of a single PP throughout the entire network. The key condition for successful transmission is that the sum of the forward and backward delays in the resonance pair matches the resonance frequency of the network modules. The transmission is much faster, by more than a factor of two, than in the original CTR mechanism. Moreover, it distinctly lowers the threshold for successful communication by synchronous spiking in modular networks of weakly coupled networks. Thus, our results suggest a new functional role of bidirectional connectivity for the communication in cortical area networks.

A New Approach to Model Predictive Control Based on Two Degrees of Freedom Control and B-Splines Input Shaping

The purpose of this article is to reduce some technical difficulties related to the complexity of stability and feasibility analysis of Model Predictive Control (MPC) as well as to reduce the complexity of the relative optimization procedure. The new approach is based on a two degrees of freedom control scheme where the output $r(k)$ of a feedforward input estimator is used as input forcing the closed-loop system $\Sigma _f$ . This latter is given by the feedback connection of a Linear Time Invariant (LTI) plant with a dynamic output controller. The task of the controller is to guarantee the stability of $\Sigma _f$ , as well as the fulfillment of hard constraints for any $r(k)$ satisfying an “ a priori ” determined admissibility condition. The input $r(k)$ is computed through the online minimization of a quadratic cost functional and is applied to $\Sigma _f$ according to the usual MPC strategy. To simplify the constrained optimization problem, the input $r(k)$ forcing $\Sigma _f$ is assumed to be given by a B-spline function. This greatly decreases the number of decision variables of the online optimization procedure because B-splines are universal approximators which admit a parsimonious parametric representation. Moreover such parameterization allows us to reformulate the minimization of the cost functional as a box Constrained Least Squares (CLS) problem. It is shown that stability and recursive feasibility of the adopted MPC strategy are guaranteed in advance, regardless the chosen prediction horizon.

Hippocampal CA1 and CA3 neural recording in the human brain: validation of depth electrode placement through high-resolution imaging and electrophysiology.

Intracranial human brain recordings typically utilize recording systems that do not distinguish individual neuron action potentials. In such cases, individual neurons are not identified by location within functional circuits. In this paper, verified localization of singly recorded hippocampal neurons within the CA3 and CA1 cell fields is demonstrated. Macro-micro depth electrodes were implanted in 23 human patients undergoing invasive monitoring for identification of epileptic seizure foci. Individual neurons were isolated and identified via extracellular action potential waveforms recorded via macro-micro depth electrodes localized within the hippocampus. A morphometric survey was performed using 3T MRI scans of hippocampi from the 23 implanted patients, as well as 46 normal (i.e., nonepileptic) patients and 26 patients with a history of epilepsy but no history of depth electrode placement, which provided average dimensions of the hippocampus along typical implantation tracks. Localization within CA3 and CA1 cell fields was tentatively assigned on the basis of recording electrode site, stereotactic positioning of the depth electrode in comparison with the morphometric survey, and postsurgical MRI. Cells were selected as candidate CA3 and CA1 principal neurons on the basis of waveform and firing rate characteristics and confirmed within the CA3-to-CA1 neural projection pathways via measures of functional connectivity. Cross-correlation analysis confirmed that nearly 80% of putative CA3-to-CA1 cell pairs exhibited positive correlations compatible with feed-forward connection between the cells, while only 2.6% exhibited feedback (inverse) connectivity. Even though synchronous and long-latency correlations were excluded, feed-forward correlation between CA3-CA1 pairs was identified in 1071 (26%) of 4070 total pairs, which favorably compares to reports of 20%-25% feed-forward CA3-CA1 correlation noted in published animal studies. This study demonstrates the ability to record neurons in vivo from specified regions and subfields of the human brain. As brain-machine interface and neural prosthetic research continues to expand, it is necessary to be able to identify recording and stimulation sites within neural circuits of interest.

Synaptic properties of the feedback connections from the thalamic reticular nucleus to the dorsal lateral geniculate nucleus.

The thalamic reticular nucleus (TRN) is a shell-like structure comprised of GABAergic neurons that surrounds the dorsal thalamus. While playing a key role in modulating thalamocortical interactions, TRN inhibition of thalamic activity is often thought of as having an all-or-none impact. Although TRN neurons have a dynamic firing range, it remains unclear how variable rates of TRN activity gate thalamocortical transmission. To address this, we examined the ultrastructural features and functional synaptic properties of the feedback connections in the mouse thalamus between TRN and the dorsal lateral geniculate nucleus (dLGN), the principal relay of retinal signals to visual cortex. Using electron microscopy to identify TRN input to dLGN, we found that TRN terminals formed synapses with non-GABAergic postsynaptic profiles. Compared with other nonretinal terminals in dLGN, those from TRN were relatively large and tended to contact proximal regions of relay cell dendrites. To evoke TRN activity in dLGN, we adopted an optogenetic approach by expressing ChR2, or a variant (ChIEF) in TRN terminals. Both in vitro and in vivo recordings revealed that repetitive stimulation of TRN terminals led to a frequency-dependent inhibition of dLGN activity, with higher rates of stimulation resulting in increasing levels of membrane hyperpolarization and corresponding decreases in spike firing. This relationship suggests that alterations in TRN activity lead to graded changes in relay cell spike firing.NEW & NOTEWORTHY The thalamic reticular nucleus (TRN) modulates thalamocortical transmission through inhibition. In mouse, TRN terminals in the dorsal lateral geniculate nucleus (dLGN) form synapses with relay neurons but not interneurons. Stimulation of TRN terminals in dLGN leads to a frequency-dependent form of inhibition, with higher rates of stimulation leading to a greater suppression of spike firing. Thus, TRN inhibition appears more dynamic than previously recognized, having a graded rather than an all-or-none impact on thalamocortical transmission.

A robust least squares based approach to min‐max model predictive control

SummaryThis article deals with the model predictive control (MPC) of linear, time‐invariant discrete‐time polytopic (LTIDP) systems. The 2‐fold aim is to simplify the treatment of complex issues like stability and feasibility analysis of MPC in the presence of parametric uncertainty as well as to reduce the complexity of the relative optimization procedure. The new approach is based on a two degrees of freedom (2DOF) control scheme, where the output r(k) of the feedforward input estimator (IE) is used as input forcing the closed‐loop system ∑f. ∑f is the feedback connection of an LTIDP plant ∑p with an LTI feedback controller ∑g. Both cases of plants with measurable and unmeasurable state are considered. The task of ∑g is to guarantee the quadratic stability of ∑f, as well as the fulfillment of hard constraints on some physical variables for any input r(k) satisfying an “a priori” determined admissibility condition. The input r(k) is computed by the feedforward IE through the on‐line minimization of a worst‐case finite‐horizon quadratic cost functional and is applied to ∑f according to the usual receding horizon strategy. The on‐line constrained optimization problem is here simplified, reducing the number of the involved constraints and decision variables. This is obtained modeling r(k) as a B‐spline function, which is known to admit a parsimonious parametric representation. This allows us to reformulate the minimization of the worst‐case cost functional as a box‐constrained robust least squares estimation problem, which can be efficiently solved using second‐order cone programming.

Minimum Cost Feedback Selection in Structured Systems: Hardness and Approximation Algorithm

This article deals with output feedback selection in linear time-invariant structured systems. We assume that the inputs and the outputs are dedicated , i.e., each input directly actuates a single state and each output directly senses a single state. Given a structured system with dedicated inputs and outputs and a cost matrix that denotes the cost of each feedback connection, our aim is to select a minimum cost set of feedback connections such that the closed-loop system satisfies arbitrary pole-placement. This problem is referred to as the optimal feedback selection problem for dedicated i/o. The optimal feedback selection problem for dedicated i/o is NP-hard and inapproximable to a constant factor of log $ n$ , where $n$ denotes the system dimension. To this end, we propose an algorithm to find an approximate solution to the problem. The proposed algorithm consists of a potential function incorporated with a greedy scheme and attains a solution with a guaranteed approximation ratio. We consider two special network topologies of practical importance, referred to as back-edge feedback structure and hierarchical networks . For the first case, which is NP-hard and inapproximable to a multiplicative factor of log $ n$ , we provide a $\text{log} n$ -approximate solution. For hierarchical networks, we give a dynamic programming based algorithm to obtain an optimal solution in polynomial time.

Role of Feedback Connections in Central Visual Processing.

The physiological response properties of neurons in the visual system are inherited mainly from feedforward inputs. Interestingly, feedback inputs often outnumber feedforward inputs. Although they are numerous, feedback connections are weaker, slower, and considered to be modulatory, in contrast to fast, high-efficacy feedforward connections. Accordingly, the functional role of feedback in visual processing has remained a fundamental mystery in vision science. At the core of this mystery are questions about whether feedback circuits regulate spatial receptive field properties versus temporal responses among target neurons, or whether feedback serves a more global role in arousal or attention. These proposed functions are not mutually exclusive, and there is compelling evidence to support multiple functional roles for feedback. In this review, the role of feedback in vision will be explored mainly from the perspective of corticothalamic feedback. Further generalized principles of feedback applicable to corticocortical connections will also be considered.

Auto-Regressive Time Delayed jump neural network for blood glucose levels forecasting

Diabetes mellitus is a widespread chronic disease and is one of the main causes of death worldwide. In order to improve the quality of life of people with diabetes and reduce the occurrence of complications, it is fundamental to prevent glycemic levels from exceeding the physiologic range. With this purpose, many works in recent years have been developed to forecast future glycemic trends using machine learning algorithms that exploit the reading of continuous glucose monitoring sensors, which gather glycemic data from diabetic patients 24 h a day. However, their application is limited in practice by the fact that they usually require a large amount of training data and other heterogeneous features gathered from patients. For this reason, in this work we present a novel neural network capable of predicting future glycemic levels using only the past glucose values as input while needing a small amount of training data. The model is a jump neural network with the addition of feedback connections from the output to the hidden layer, and time delays for each of the input-to-hidden, output-to-hidden and input-to-output connections. Experiments were conducted on a private and a public dataset. We evaluated performance in terms of RMSE and of adverse event detection. The proposed model outperforms other methods suited for time series forecasting, as well as models for blood glucose level prediction present in the literature.

Two-Path Network with Feedback Connections for Pan-Sharpening in Remote Sensing

High-resolution multi-spectral images are desired for applications in remote sensing. However, multi-spectral images can only be provided in low resolutions by optical remote sensing satellites. The technique of pan-sharpening wants to generate high-resolution multi-spectral (MS) images based on a panchromatic (PAN) image and the low-resolution counterpart. The conventional deep learning based pan-sharpening methods process the panchromatic and the low-resolution image in a feedforward manner where shallow layers fail to access useful information from deep layers. To make full use of the powerful deep features that have strong representation ability, we propose a two-path network with feedback connections, through which the deep features can be rerouted for refining the shallow features in a feedback manner. Specifically, we leverage the structure of a recurrent neural network to pass the feedback information. Besides, a power feature extraction block with multiple projection pairs is designed to handle the feedback information and to produce power deep features. Extensive experimental results show the effectiveness of our proposed method.

A trusted medical image super-resolution method based on feedback adaptive weighted dense network.

High-resolution (HR) medical images are preferred in clinical diagnoses and subsequent analysis. However, the acquisition of HR medical images is easily affected by hardware devices. As an effective and trusted alternative method, the super-resolution (SR) technology is introduced to improve the image resolution. Compared with traditional SR methods, the deep learning-based SR methods can obtain more clear and trusted HR images. In this paper, we propose a trusted deep convolutional neural network-based SR method named feedback adaptive weighted dense network (FAWDN) for HR medical image reconstruction. Specifically, the proposed FAWDN can transmit the information of the output image to the low-level features by a feedback connection. To explore advanced feature representation and reduce the feature redundancy in dense blocks, an adaptive weighted dense block (AWDB) is introduced to adaptively select the informative features. Experimental results demonstrate that our FAWDN outperforms the state-of-the-art image SR methods and can obtain more clear and trusted medical images than comparative methods.

Timing of response onset and offset in macaque V4: stimulus and task dependence.

In the primate visual cortex, both the magnitude of the neuronal response and its timing can carry important information about the visual world, but studies typically focus only on response magnitude. Here, we examine the onset and offset latency of the responses of neurons in area V4 of awake, behaving macaques across several experiments in the context of a variety of stimuli and task paradigms. Our results highlight distinct contributions of stimuli and tasks to V4 response latency. We found that response onset latencies are shorter than typically cited (median = 75.5 ms), supporting a role for V4 neurons in rapid object and scene recognition functions. Moreover, onset latencies are longer for smaller stimuli and stimulus outlines, consistent with the hypothesis that longer latencies are associated with higher spatial frequency content. Strikingly, we found that onset latencies showed no significant dependence on stimulus occlusion, unlike in inferotemporal cortex, nor on task demands. Across the V4 population, onset latencies had a broad distribution, reflecting the diversity of feedforward, recurrent, and feedback connections that inform the responses of individual neurons. Response offset latencies, on the other hand, displayed the opposite tendency in their relationship to stimulus and task attributes: they are less influenced by stimulus appearance but are shorter in guided saccade tasks compared with fixation tasks. The observation that response latency is influenced by stimulus- and task-associated factors emphasizes a need to examine response timing alongside firing rate in determining the functional role of area V4.NEW & NOTEWORTHY Onset and offset timing of neuronal responses can provide information about visual environment and neuron's role in visual processing and its anatomical connectivity. In the first comprehensive examination of onset and offset latencies in the intermediate visual cortical area V4, we find neurons respond faster than previously reported, making them ideally suited to contribute to rapid object and scene recognition. While response onset reflects stimulus characteristics, timing of response offset is influenced more by behavioral task.

Backpropagation and the brain.

During learning, the brain modifies synapses to improve behaviour. In the cortex, synapses are embedded within multilayered networks, making it difficult to determine the effect of an individual synaptic modification on the behaviour of the system. The backpropagation algorithm solves this problem in deep artificial neural networks, but historically it has been viewed as biologically problematic. Nonetheless, recent developments in neuroscience and the successes of artificial neural networks have reinvigorated interest in whether backpropagation offers insights for understanding learning in the cortex. The backpropagation algorithm learns quickly by computing synaptic updates using feedback connections to deliver error signals. Although feedback connections are ubiquitous in the cortex, it is difficult to see how they could deliver the error signals required by strict formulations of backpropagation. Here we build on past and recent developments to argue that feedback connections may instead induce neural activities whose differences can be used to locally approximate these signals and hence drive effective learning in deep networks in the brain.

Predictive processing account of action perception: Evidence from effective connectivity in the action observation network

Visual perception of actions is supported by a network of brain regions in the occipito-temporal, parietal, and premotor cortex in the primate brain, known as the Action Observation Network (AON). Although there is a growing body of research that characterizes the functional properties of each node of this network, the communication and direction of information flow between the nodes is unclear. According to the predictive coding account of action perception (Kilner, Friston, & Frith, 2007a; 2007b), this network is not a purely feedforward system but has backward connections through which prediction error signals are communicated between the regions of the AON. In the present study, we investigated the effective connectivity of the AON in an experimental setting where the human subjects' predictions about the observed agent were violated, using fMRI and Dynamical Causal Modeling (DCM). We specifically examined the influence of the lowest and highest nodes in the AON hierarchy, pSTS and ventral premotor cortex, respectively, on the middle node, inferior parietal cortex during prediction violation. Our DCM results suggest that the influence on the inferior parietal node is through a feedback connection from ventral premotor cortex during perception of actions that violate people's predictions.