Search bioRxiv⌕ Search

Biology subjects

Meffin, H.

Publications and source records attributed to Meffin, H..

9 recordsLinked to original sources
bioRxiv2025

Modelling Predictive Coding in the Primary Visual Cortex (V1): Layer 2/3 Circuits for Prediction Error Computation through Compartmentalized Spiking Neurons

Cortical Layer 2/3 has been consistently implicated as the locus of prediction-error signalling in hierarchical models of cortical sensory processing. However, the circuit mechanisms that generate biologically plausible prediction-error (PE) signals remain elusive. A spiking network model is presented here in which two-compartment excitatory pyramidal neurons interact with three inhibitory subtypes: parvalbumin-expressing (PV), somatostatin-expressing (SOM), and vasoactive-intestinal-peptide-expressing (VIP) interneurons, to compute sign-specific prediction errors (positive and negative PEs). Feedforward input targets the soma, whereas top-down feedback reaches the distal apical dendrite, enabling a local somato-dendritic comparison. A PE emerges whenever the balance between excitation and inhibition is selectively disrupted within one compartment, recruiting either positive-error (PE+) or negative-error (PE-) subpopulations of pyramidal neurons. Unlike prior learning-dependent, rate-based accounts, this fixed-weight spiking circuit shows that bidirectional PE signals (PE+ and PE-) can arise online from compartment-specific balance without any synaptic weight updates. The model reproduces key experimental observations, including sparse mismatch responses, compartment-specific inhibition, and VIP-mediated disinhibition. Across four canonical sensory-prediction configurations, the circuit maintains a tight balance during matched input and generates bidirectional PE signals only under mismatch. By routing sensory drive from Layer 4 into Layer 2/3 and allowing the resulting PE activity to project toward deeper feedback generators, the model situates Layer 2/3 as a dedicated, feature-specific mismatch detector within a hierarchical inference network. These results provide a mechanistic bridge from dendritic computation to laminar predictive coding, demonstrating how realistic spiking dynamics can implement fast, sign-specific PE signaling without learning. Author summaryIn this study, we present a biologically grounded spiking model of layer 2/3 in primary visual cortex within the predictive coding framework. Our goal is to explain how superficial cortical circuits compute fast, sign-specific prediction errors when sensory input does not match top-down expectations. The model uses two-compartment pyramidal neurons whose somata receive feedforward drive from layer 4 while apical dendrites receive feedback, together with three key inhibitory interneuron classes, parvalbumin-expressing (PV), somatostatin-expressing (SOM), and vasoactive-intestinal-peptide-expressing (VIP), that provide compartment-specific inhibition and disinhibition. When input and prediction match, excitation and inhibition remain tightly balanced and activity is sparse; when they differ, this balance is transiently broken in the appropriate compartment, and distinct populations signal either a positive error (unexpected presence) or a negative error (unexpected absence). The circuit reproduces several in vivo observations in layer 2/3, including sparse mismatch responses, compartment-specific inhibition, and VIP-mediated disinhibition, and it does so without requiring synaptic weight changes. By routing feature-selective signals from layer 4 into layer 2/3 and relaying the resulting errors toward deeper layers, the model positions layer 2/3 as a local, feature-specific mismatch detector in a hierarchical system. This work provides a concrete, testable mechanism linking dendritic computation, inhibitory diversity, and predictive coding in the cortex.

neuroscience↗
bioRxiv2024

Emergence of Sparse Coding, Balance and Decorrelation from a Biologically-Grounded Spiking Neural Network Model of Learning in the Primary Visual Cortex

1Many computational studies attempt to address the question of information representation in biological neural networks using an explicit optimization based on an objective function. These approaches begin with principles of information representation that are expected to be found in the network and from which learning rules can be derived. This study approaches the question from the opposite direction; beginning with a model built upon the experimentally observed properties of neural responses, homeostasis, and synaptic plasticity. The known properties of information representation are then expected to emerge from this substrate. A spiking neural model of the primary visual cortex (V1) was investigated. Populations of both inhibitory and excitatory leaky integrate-and-fire neurons with recurrent connections were provided with spiking input from simulated ON and OFF neurons of the lateral geniculate nucleus. This network was provided with natural image stimuli as input. All synapses underwent learning using spike-timing-dependent plasticity learning rules. A homeostatic rule adjusted the weights and thresholds of each neuron based on target homeostatic spiking rates and mean synaptic input values. These experimentally grounded rules resulted in a number of the expected properties of information representation. The network showed a temporally sparse spike response to inputs and this was associated with a sparse code with Gabor-like receptive fields. The network was balanced at both slow and fast time scales; increased excitatory input was balanced by increased inhibition. This balance was associated with decorrelated firing that was observed as population sparseness. This population sparseness was both the cause and result of the decorrelation of receptive fields. These observed emergent properties (balance, temporal sparseness, population sparseness, and decorrelation) indicate that the network is implementing expected principles of information processing: efficient coding, information maximization (infomax), and a lateral or single-layer form of predictive coding. These emergent features of the network were shown to be robust to randomized jitter of the values of key simulation parameters.

neuroscience↗
bioRxiv2023

Neural Activity Shaping in Visual Prostheses with Deep Learning

ObjectiveThe visual perception provided by retinal prostheses is poor and limited to images constructed of phosphenes generated by the electrodes. One limiting factor has been the conventional strategy used to encode the target image into a stimulation pattern. Under this strategy, if the electrode density is high, the current spread of neighbouring unipolar stimuli overlaps, leading to blurred images. Simultaneous multipolar stimulation guided by the measured neural responses can attenuate excessive spread of excitation and allows for a more precise electrical input to the retina. However, it is far from trivial to predict what multipolar stimulus pattern will elicit the desired retinal response for a given target image. Here, we propose to solve this problem using an Artificial Neural Network (ANN) that could be trained with data acquired from the implant itself. ApproachOur method consists of two ANNs trained sequentially. The Measurement Predictor Network (MPN) is trained on data from the implant and is used to predict how the retina responds to multipolar stimulation by learning the forward model. The Stimulus Generator Network (STG) is trained on a large dataset of natural images and uses the trained MPN to determine efficient multipolar stimulus patterns by learning the inverse model. We validate our method in silico using a realistic model of retinal response to multipolar stimulation. Main ResultsWe show that the simulated retinal activations elicited with our ANN-based approach are considerably sharper when compared with the conventional method used in existing devices. The SGN finds multipolar stimulation patterns that are tuned to a specific retina, thus providing patient-specific stimuli. Also, due to its small computational cost, the SGN can output stimulation patterns at a very high rate. SignificanceOur novel protocol opens the door to personalized multipolar retinal stimulation, which may improve the visual experience and quality of life of retinal prosthesis users.

bioengineering↗
bioRxiv2022

Quantifying Visual Acuity for Pre-Clinical Testing of Visual Prostheses.

Visual prostheses currently restore only limited vision. More research and pre-clinical work are required to improve the devices and stimulation strategies that are used to induce neural activity that results in visual perception. Evaluation of candidate strategies and devices requires an objective way to convert measured and modelled patterns of neural activity into a quantitative measure of visual acuity. This study presents an approach that compares evoked patterns of neural activation with target and reference patterns. A d-prime measure of discriminability determines whether the evoked neural activation pattern is sufficient to discriminate between the target and reference patterns and thus provide a quantified level of visual perception in the clinical Snellen and MAR scales. The measure was accurate in providing an estimate of the perceivable feature sizes in scaled standardized "C" and "E" optotypes. The approach was used to assess the visual acuity provided by two alternative stimulation strategies applied to simulated retinal implants with different phosphene sizes and electrode pitch configurations. It was found that when there is substantial overlap in neural activity generated by different electrodes, an estimate of acuity based only upon electrode pitch is incorrect; our proposed method gives an accurate result in these circumstances. Quantification of visual acuity using this approach in pre-clinical development will allow for more rapid and accurate prototyping of improved devices and neural stimulation strategies.

bioengineering↗
bioRxiv2021

How Stimulus Statistics Affect The Receptive Fields of V1 Cells

We studied the changes that neuronal RF models undergo when the statistics of the stimulus are changed from those of white Gaussian noise (WGN) to those of natural scenes (NS). Fitting the model to data estimates both a cascade of linear filters on the stimulus, as wells as the static nonlinearities that map the output of the filters to the neuronal spike rates. We found that cells respond differently to these two classes of stimuli, with mostly higher spike rates and shorter response latencies to NS than to WGN. The most striking finding was that NS resulted in RFs that had additional uncovered filters than did WGN. This finding was not an artefact of the higher spike rates but rather related to a change in coding. Our results reveal a greater extent of nonlinear processing in V1 neurons when stimulated using NS compared to WGN. Our findings indicate the existence of nonlinear mechanisms that endow V1 neurons with context-dependent transmission of visual information.

neuroscience↗
bioRxiv2020

Learning receptive field properties of complex cells in V1

There are two distinct classes of cells in the primary visual cortex (V1): simple cells and complex cells. One defining feature of complex cells is their spatial phase invariance; they respond strongly to oriented grating stimuli with a preferred orientation but with a wide range of spatial phases. A classical model of complete spatial phase invariance in complex cells is the energy model, in which the responses are the sum of the squared outputs of two linear spatially phase-shifted filters. However, recent experimental studies have shown that complex cells have a diverse range of spatial phase invariance and only a subset can be characterized by the energy model. While several models have been proposed to explain how complex cells could learn to be selective to orientation but invariant to spatial phase, most existing models overlook many biologically important details. We propose a biologically plausible model for complex cells that learns to pool inputs from simple cells based on the presentation of natural scene stimuli. The model is a three-layer network with rate-based neurons that describes the activities of LGN cells (layer 1), V1 simple cells (layer 2), and V1 complex cells (layer 3). The first two layers implement a recently proposed simple cell model that is biologically plausible and accounts for many experimental phenomena. The neural dynamics of the complex cells is modeled as the integration of simple cells inputs along with response normalization. Connections between LGN and simple cells are learned using Hebbian and anti-Hebbian plasticity. Connections between simple and complex cells are learned using a modified version of the Bienenstock, Cooper, and Munro (BCM) rule. Our results demonstrate that the learning rule can describe a diversity of complex cells, similar to those observed experimentally. Author summaryMany cortical functions originate from the learning ability of the brain. How the properties of cortical cells are learned is vital for understanding how the brain works. There are many models that explain how V1 simple cells can be learned. However, how V1 complex cells are learned still remains unclear. In this paper, we propose a model of learning in complex cells based on the Bienenstock, Cooper, and Munro (BCM) rule. We demonstrate that properties of receptive fields of complex cells can be learned using this biologically plausible learning rule. Quantitative comparisons between the model and experimental data are performed. Results show that model complex cells can account for the diversity of complex cells found in experimental studies. In summary, this study provides a plausible explanation for how complex cells can be learned using biologically plausible plasticity mechanisms. Our findings help us to better understand biological vision processing and provide us with insights into the general signal processing principles that the visual cortex employs to process visual information.

neuroscience↗
bioRxiv2020

Mechanisms of Feature Selectivity and Invariance in Primary Visual Cortex

Visual object identification requires both selectivity for specific visual features that are important to the objects identity and invariance to feature manipulations. For example, a hand can be shifted in position, rotated, or contracted but still be recognised as a hand. How are the competing requirements of selectivity and invariance built into the early stages of visual processing? Typically, cells in the primary visual cortex are classified as either simple or complex. They both show selectivity for edge-orientation but complex cells develop invariance to edge position within the receptive field (spatial phase). Using a data-driven model that extracts the spatial structures and nonlinearities associated with neuronal computation, we show that the balance between selectivity and invariance in complex cells is more diverse than thought. Phase invariance is frequently partial, thus retaining sensitivity to brightness polarity, while invariance to orientation and spatial frequency are more extensive than expected. The invariance arises due to two independent factors: (1) the structure and number of filters and (2) the form of nonlinearities that act upon the filter outputs. Both vary more than previously considered, so primary visual cortex forms an elaborate set of generic feature sensitivities, providing the foundation for more sophisticated object processing.

neuroscience↗
bioRxiv2018

Minimizing activation of overlying axons with epiretinal stimulation: The role of fiber orientation and electrode configuration

AO_SCPCAPBSTRACTC_SCPCAPObjective. Currently, a challenge in electrical stimulation of the retina is to excite only the cells lying directly under the electrode in the ganglion cell layer, while avoiding excitation of the axons that pass over the surface of the retina in the nerve fiber layer. Since these passing fibers may originate from distant regions of the ganglion cell layer. Stimulation of both target retinal ganglion cells and overlying axons results in irregular visual percepts, significantly limiting perceptual efficacy. This research explores how differences in fiber orientation between the nerve fiber layer and ganglion cell layer leads to differences in the activation of the axon initial segment and axons of passage. Approach. Axons of passage of retinal ganglion cells in the nerve fiber layer are characterized by a narrow distribution of fiber orientations, causing highly anisotropic spread of applied current. In contrast, proximal axons in the ganglion cell layer have a wider distribution of orientations. A four-layer computational model of epiretinal extracellular stimulation that captures the effect of neurite orientation in anisotropic tissue has been developed using a modified version of the standard volume conductor model, known as the cellular composite model. Simulations are conducted to investigate the interaction of neural tissue orientation, stimulating electrode configuration, and stimulation pulse duration and amplitude. Main results. The dependence of fiber activation on the anisotropic nature of the nerve fiber layer is first established. Via a comprehensive search of key parameters, our model shows that the simultaneous stimulation with multiple electrodes aligned with the nerve fiber layer can be used to achieve selective activation of axon initial segments rather than passing fibers. This result can be achieved with only a slight increase in total stimulus current and modest increases in the spread of activation in the ganglion cell layer, and is shown to extend to the general case of arbitrary electrode array positioning and arbitrary target neural volume. Significance. These results elucidate a strategy for more targeted stimulation of retinal ganglion cells with experimentally-relevant multi-electrode geometries and readily achievable stimulation requirements.

biophysics↗