Biology subjects

Wessel, R.

Publications and source records attributed to Wessel, R..

11 recordsLinked to original sources

bioRxiv2025

The representational geometry of out-of-distribution generalization in primary visual cortex and artificial neural networks

Humans and other animals display a remarkable ability to generalize learned knowledge to novel domains, a phenomenon known as out-of-distribution (OOD) generalization. This capability is thought to depend on the format of neural population representations; however, the specific geometrical properties that support OOD generalization and the learning objectives that give rise to them remain poorly understood. Here, we examine the OOD generalization of neural population representations of static grating orientations in the mouse visual cortex. We show that a decoder trained on neural responses within a restricted orientation domain can generalize to held-out orientation domains. The quality of generalization correlates with both the dimensionality and the curvature of the underlying neural representation manifold. Notably, similar OOD-generalizable geometry emerges in a deep neural network trained to predict the next frame in natural video sequences. These findings reveal the representational geometric properties underlying OOD generalization, and suggest that predictive learning objectives offer a promising approach for acquiring generalizable representation geometry.

neuroscience↗

bioRxiv2024

Speed modulations in grid cell information geometry

Grid cells, known for their hexagonal spatial firing patterns, are widely regarded as essential to the brains internal representation of the external space. Maintaining an accurate internal spatial representation is challenging when an animal is running at high speeds, as its self-location constantly changes. Previous studies of speed modulation of grid cells focused on individual or pairs of grid cells, yet neurons represent information via collective population activity. Population noise covariance can have significant impact on information coding that is impossible to infer from individual neuron analysis. To address this issue, we developed a novel Gaussian Process with Kernel Regression (GKR) method that allows study the simultaneously recorded neural population representation from an information geometry framework. We applied GKR to grid cell population activity, and found that running speed increases both grid cell activity toroidal-like manifold size and noise strength. Importantly, the effect of manifold dilation outpaces the effect of noise increasement, as indicated by the overall higher Fisher information at increasing speeds. This result is further supported by improved spatial information decoding accuracy at high speeds. Finally, we showed that the existence of noise covariance is information detrimental because it causes more noise projected onto the manifold surface. In total, our results indicate that grid cell spatial coding improves with increasing running speed. GKR provides a useful tool to understand neural population coding from an intuitive information geometric perspective.

neuroscience↗

bioRxiv2024

Brain-like border ownership signals support prediction of natural videos

To make sense of visual scenes, the brain must segment foreground from background. This is thought to be facilitated by neurons in the primate visual system that encode border ownership (BOS), i.e. whether a local border is part of an object on one or the other side of the border. It is unclear how these signals emerge in neural networks without a teaching signal of what is foreground and background. In this study, we investigated whether BOS signals exist in PredNet, a self-supervised artificial neural network trained to predict the next image frame of natural video sequences. We found that a significant number of units in PredNet are selective for BOS. Moreover these units share several other properties with the BOS neurons in the brain, including robustness to scene variations that constitute common object transformations in natural videos, and hysteresis of BOS signals. Finally, we performed ablation experiments and found that BOS units contribute more to prediction than non-BOS units for videos with moving objects. Our findings indicate that BOS units are especially useful to predict future input in natural videos, even when networks are not required to segment foreground from background. This suggests that BOS neurons in the brain might be the result of evolutionary or developmental pressure to predict future input in natural, complex dynamic visual environments.

neuroscience↗

bioRxiv2023

Tauopathy severely disrupts homeostatic set-points in emergent neural dynamics but not the activity of individual neurons.

The homeostatic regulation of neuronal activity is essential for robust computation; key set-points, such as firing rate, are actively stabilized to compensate for perturbations. From this perspective, the disruption of brain function central to neurodegenerative disease should reflect impairments of computationally essential set-points. Despite connecting neurodegeneration to functional outcomes, the impact of disease on set-points in neuronal activity is unknown. Here we present a comprehensive, theory-driven investigation of the effects of tau-mediated neurodegeneration on homeostatic set-points in neuronal activity. In a mouse model of tauopathy, we examine 27,000 hours of hippocampal recordings during free behavior throughout disease progression. Contrary to our initial hypothesis that tauopathy would impact set-points in spike rate and variance, we found that cell-level set-points are resilient to even the latest stages of disease. Instead, we find that tauopathy disrupts neuronal activity at the network-level, which we quantify using both pairwise measures of neuron interactions as well as measurement of the networks nearness to criticality, an ideal computational regime that is known to be a homeostatic set-point. We find that shifts in network criticality 1) track with symptoms, 2) predict underlying anatomical and molecular pathology, 3) occur in a sleep/wake dependent manner, and 4) can be used to reliably classify an animals genotype. Our data suggest that the critical set-point is intact, but that homeostatic machinery is progressively incapable of stabilizing hippocampal networks, particularly during waking. This work illustrates how neurodegenerative processes can impact the computational capacity of neurobiological systems, and suggest an important connection between molecular pathology, circuit function, and animal behavior.

neuroscience↗

bioRxiv2022

Sleep restores an optimal computational regime in cortical networks

SUMMARYSleep is vitally important for brain function, yet its core restorative process remains an open question. Sleep is assumed to subserve homeostatic processes in the brain because sleep restores functional capacity, and stable function requires compensatory tuning of circuits in the face of experience. However, the set-point around which sleep tunes circuit computation is unknown; for more than four decades, the homeostatic aspect of sleep has been approximated by a heuristic model whose strongest correlate is Slow-wave Activity (SWA). While SWA can indicate sleep pressure, it fails to explain why animals need sleep. In contrast, criticality is a computational regime that optimizes information processing capacity, and is a homeostatically regulated set-point in isocortical circuits. Consistent with the effects of waking, criticality is degraded by experience-dependent plasticity. Whether criticality is the computational set-point of sleep is unknown. To address this question, we evaluated the effects of sleep and wake on emergent dynamics in ensembles of cortical neurons recorded continuously for 10-14 d in freely behaving rats. We show that normal waking experience progressively disrupts criticality, and that sleep functions to restore critical dynamics. Criticality is perturbed in a context-dependent manner depending on behavior and environmental variables, and waking experience is causal in driving these effects. The degree of deviation from criticality is robustly predictive of future sleep/wake behavior, more accurate than SWA, behavioral history, and other neural measures. Our results demonstrate that perturbation and recovery of criticality is a network homeostatic mechanism consistent with the core, restorative function of sleep.

neuroscience↗

bioRxiv2018

Critical dynamics are a homeostatic set point of cortical networks in vivo.

The brain is constantly challenged by severely destabilizing forces: proteins turn over rapidly, Hebbian modifications alter and introduce positive feedback into networks, and environments change over many timescales. Homeostatic plasticity mechanisms, which operate via negative feedback, are believed to compensate for these changes and constrain neuronal activity to a firing rate (FR) set point1,2,3. For decades, it has been widely assumed that activity in neural networks is robust as a direct result of the widespread expression of FR homeostasis1,4. Here we reveal that network dynamics are stabilized independent of excitatory FR homeostasis and that cortical networks actively self-organize around an ideal computational regime.\n\nWe continuously tracked network spiking activity in the visual cortex (V1) of freely behaving rats for nine days. We found that, under baseline conditions, networks of excitatory neurons are robustly organized around criticality, a regime known to maximize information capacity and dynamic range. Monocular deprivation (MD) revealed a dissociation of excitatory FRs and network dynamics. MD immediately and severely disrupted network organization, which returned precisely to criticality over 48h. In contrast, both the excitatory FR drop and the subsequent FR recovery trailed the timecourse of network changes by more than 30h. Model investigations suggest a role for inhibitory neurons in maintaining critical dynamics. Collectively, these results show that complex activity in cortical circuits is actively maintained near criticality and that this organization is not explained by previously identified mechanisms of pyramidal neuron FR homeostasis.

neuroscience↗

bioRxiv2018

Single-cell membrane potential fluctuations evince network scale-freeness and quasicriticality

What information single neurons receive about general neural circuit activity is a fundamental question for neuroscience. Somatic membrane potential fluctuations are driven by the convergence of synaptic inputs from a diverse cross-section of upstream neurons. Furthermore, neural activity is often scale-free implying that some measurements should be the same, whether taken at large or small scales. Together, convergence and scale-freeness support the hypothesis that single membrane potential recordings carry useful information about high-dimensional cortical activity. Conveniently, the theory of \"critical branching networks\" (a purported explanation for scale-freeness) provides testable predictions about scale-free measurements which are readily applied to membrane potential fluctuations. To investigate, we obtained whole-cell current clamp recordings of pyramidal neurons in visual cortex of turtles with unknown genders. We isolated fluctuations in membrane potential below the firing threshold and analyzed them by adapting the definition of \"neuronal avalanches\" (spurts of population spiking). The membrane potential fluctuations we analyzed were scale-free and consistent with critical branching. These findings recapitulated results from large-scale cortical population data obtained separately in complementary experiments using microelectrode arrays (previously published (Shew et al., 2015)). Simultaneously recorded single-unit local field potential did not provide a good match; demonstrating the specific utility of membrane potential. Modeling shows that estimation of dynamical network properties from neuronal inputs is most accurate when networks are structured as critical branching networks. In conclusion, these findings extend evidence for critical branching while also establishing subthreshold pyramidal neuron membrane potential fluctuations as an informative gauge of high-dimensional cortical population activity.

neuroscience↗

bioRxiv2017

Visual Response Properties In The Three Layer Turtle Visual Cortex

Turtle dorsal cortex provides us with unique insights into cortical processing. It is known to share many features with the mammalian hippocampus and olfactory cortex as well as geniculo-cortical areas in stem amniotes from which mammals evolved. To this end, we have used data from extracellular recordings from microelectrode arrays to study spatial and temporal patterns of responses to visual stimuli as seen in both local field potential and action potentials. We discovered surprisingly large receptive fields, responsiveness to a broad range of stimuli, and high correlation between distant neural ensembles across recording array. Moreover, we found significant response variability regarding latency and strength in the presence of adaptation to both ongoing and visually evoked activity.

neuroscience↗

bioRxiv2016

Network activity influences the subthreshold and spiking visual responses of pyramidal neurons in a three-layer cortex

A primary goal of systems neuroscience is to understand cortical function, which typically involves studying spontaneous and sensory-evoked cortical activity. Mounting evidence suggests a strong and complex relationship between the ongoing and evoked state. To date, most work in this area has been based on spiking in populations of neurons. While advantageous in many respects, this approach is limited in scope; it records the activities of a minority of neurons, and gives no direct indication of the underlying subthreshold dynamics. Membrane potential recordings can fill these gaps in our understanding, but are difficult to obtain in vivo. Here, we record subthreshold cortical visual responses in the ex vivo turtle eye-attached whole-brain preparation, which is ideally-suited to such a study. In the absence of visual stimulation, the network is \"synchronous\"; neurons display network-mediated transitions between low- and high-conductance membrane potential states. The prevalence of these slow-wave transitions varies across turtles and recording sessions. Visual stimulation evokes similar high-conductance states, which are on average larger and less reliable when the ongoing state is more synchronous. Responses are muted when immediately preceded by large, spontaneous high-conductance events. Evoked spiking is sparse, highly variable across trials, and mediated by concerted synaptic inputs that are in general only very weakly correlated with inputs to nearby neurons. Together, these results highlight the multiplexed influence of the cortical network on the spontaneous and sensory-evoked activity of individual cortical neurons.

neuroscience↗

bioRxiv2016

Response variability and population coupling of cortical synaptic inputs are strongly influenced by network properties

The highly variable spiking of a cortical neuron is \"coupled\" to that of other neurons in the network. This has implications for sensory coding, and appears to represent a fundamental property of cortical sensory processing. To date, most studies of population coupling have focused on recorded spiking activity, an approach that suffers from several confounding issues. Moreover, the contributions of various network properties to population coupling are largely unexplored. To this end, we recorded the membrane potential (V) and the nearby LFP in the visual cortex of the turtle ex vivo wholebrain preparation during ongoing and visually-evoked activity. We used an algorithm to infer the excitatory conductance (g) from V, and calculated the g-LFP coupling. We found that g-LFP coupling was highly variable across neurons, and increased following visual stimulation before relaxing to intermediate values. To investigate the role of the network, we implemented a driven small-world network of leaky integrate-and-fire neurons. This model reproduces the large across-trial response variability and g-LFP coupling dynamic, and suggests crucial roles for anatomical and emergent network properties.

neuroscience↗

bioRxiv2016

Adaptation modulates correlated response variability in visual cortex

Cortical sensory responses are highly variable across stimulus presentations. This variability can be correlated across neurons (due to some combination of dense intracortical connectivity, cortical activity level, and cortical state), with fundamental implications for population coding. Yet the interpretation of correlated response variability (or \"noise correlation\") has remained fraught with difficulty, in part because of the restriction to extracellular neuronal spike recordings. Here, we measured response variability and its correlation at the most microscopic level of electrical neural activity, the membrane potential, by obtaining dual whole-cell recordings from pairs of cortical pyramidal neurons during visual processing. We found that during visual stimulation, correlated variability adapts towards an intermediate level and that this correlation dynamic is mediated by intracortical mechanisms. A model network with external inputs, synaptic depression, and structure reproduced the observed dynamics of correlated variability. These results establish that intracortical adaptation self-organizes cortical circuits towards a balanced regime at which network coordination maintains an intermediate level.

neuroscience↗