Random Compressed Coding with Neurons

Classical models of efficient coding in neurons assume simple mean responses--'tuning curves'--such as bell shaped or monotonic functions of a stimulus feature. Real neurons, however, can be more complex: grid cells, for example, exhibit periodic responses which impart the neural population code with high accuracy. But do highly accurate codes require fine tuning of the response properties? We address this question with the use of a benchmark model: a neural network with random synaptic weights which result in output cells with irregular tuning curves. Irregularity enhances the local resolution of the code but gives rise to catastrophic, global errors. For optimal smoothness of the tuning curves, when local and global errors balance out, the neural network compresses information from a high-dimensional representation to a low-dimensional one, and the resulting distributed code achieves exponential accuracy. An analysis of recordings from monkey motor cortex points to such 'compressed efficient coding'. Efficient codes do not require a finely tuned design--they emerge robustly from irregularity or randomness.

Finite Element Modeling of Magnitude and Location of Brain Micromotion Induced Strain for Intracortical Implants

Micromotion-induced stress remains one of the main determinants of life of intracortical implants. This is due to high stress leading to tissue injury, which in turn leads to an immune response coupled with a significant reduction in the nearby neural population and subsequent isolation of the implant. In this work, we develop a finite element model of the intracortical probe-tissue interface to study the effect of implant micromotion, implant thickness, and material properties on the strain levels induced in brain tissue. Our results showed that for stiff implants, the strain magnitude is dependent on the magnitude of the motion, where a micromotion increase from 1 to 10 μm induced an increase in the strain by an order of magnitude. For higher displacement over 10 μm, the change in the strain was relatively smaller. We also showed that displacement magnitude has no impact on the location of maximum strain and addressed the conflicting results in the literature. Further, we explored the effect of different probe materials [i.e., silicon, polyimide (PI), and polyvinyl acetate nanocomposite (PVAc-NC)] on the magnitude, location, and distribution of strain. Finally, we showed that strain distribution across cortical implants was in line with published results on the size of the typical glial response to the neural probe, further reaffirming that strain can be a precursor to the glial response.

Similar neural and perceptual masking effects of low-power optogenetic stimulation in primate V1

Can direct stimulation of primate V1 substitute for a visual stimulus and mimic its perceptual effect? To address this question, we developed an optical-genetic toolkit to 'read' neural population responses using widefield calcium imaging, while simultaneously using optogenetics to 'write' neural responses into V1 of behaving macaques. We focused on the phenomenon of visual masking, where detection of a dim target is significantly reduced by a co-localized medium-brightness mask [1, 2]. Using our toolkit, we tested whether V1 optogenetic stimulation can recapitulate the perceptual masking effect of a visual mask. We find that, similar to a visual mask, low-power optostimulation can significantly reduce visual detection sensitivity, that a sublinear interaction between visual and optogenetic evoked V1 responses could account for this perceptual effect, and that these neural and behavioral effects are spatially selective. Our toolkit and results open the door for further exploration of perceptual substitutions by direct stimulation of sensory cortex.

Behavioral Prioritization Enhances Working Memory Precision and Neural Population Gain

Humans allocate visual working memory (WM) resource according to behavioral relevance, resulting in more precise memories for more important items. Theoretically, items may be maintained by feature-tuned neural populations, where the relative gain of the populations encoding each item determines precision. To test this hypothesis, we compared the amplitudes of delay period activity in the different parts of retinotopic maps representing each of several WM items, predicting the amplitudes would track behavioral priority. Using fMRI, we scanned participants while they remembered the location of multiple items over a WM delay and then reported the location of one probed item using a memory-guided saccade. Importantly, items were not equally probable to be probed (0.6, 0.3, 0.1, 0.0), which was indicated with a precue. We analyzed fMRI activity in 10 visual field maps in occipital, parietal, and frontal cortex known to be important for visual WM. In early visual cortex, but not association cortex, the amplitude of BOLD activation within voxels corresponding to the retinotopic location of visual WM items increased with the priority of the item. Interestingly, these results were contrasted with a common finding that higher-level brain regions had greater delay period activity, demonstrating a dissociation between the absolute amount of activity in a brain area and the activity of different spatially selective populations within it. These results suggest that the distribution of WM resources according to priority sculpts the relative gains of neural populations that encode items, offering a neural mechanism for how prioritization impacts memory precision.

The interaction between neural populations: Additive versus diffusive coupling

Models of networks of populations of neurons commonly assume that the interactions between neural populations are via additive or diffusive coupling. When using the additive coupling, a population's activity is affected by the sum of the activities of neighbouring populations. In contrast, when using the diffusive coupling a neural population is affected by the sum of the differences between its activity and the activity of its neighbours. These two coupling functions have been used interchangeably for similar applications. Here, we show that the choice of coupling can lead to strikingly different brain network dynamics. We focus on a model of seizure transitions that has been used both with additive and diffusive coupling in the literature. We consider networks with two and three nodes, and large random and scale-free networks with 64 nodes. We further assess functional networks inferred from magnetoencephalography (MEG) from people with epilepsy and healthy controls. To characterize the seizure dynamics on these networks, we use the escape time, the brain network ictogenicity (BNI) and the node ictogenicity (NI), which are measures of the network's global and local ability to generate seizures. Our main result is that the level of ictogenicity of a network is strongly dependent on the coupling function. We find that people with epilepsy have higher additive BNI than controls, as hypothesized, while the diffusive BNI provides the opposite result. Moreover, individual nodes that are more likely to drive seizures with one type of coupling are more likely to prevent seizures with the other coupling function. Our results on the MEG networks and evidence from the literature suggest that the additive coupling may be a better modelling choice than the diffusive coupling, at least for BNI and NI studies. Thus, we highlight the need to motivate and validate the choice of coupling in future studies.

Quasi-universal scaling in mouse-brain neuronal activity stems from edge-of-instability critical dynamics

The brain is in a state of perpetual reverberant neural activity, even in the absence of specific tasks or stimuli. Shedding light on the origin and functional significance of such activity is essential to understanding how the brain transmits, processes, and stores information. An inspiring, albeit controversial, conjecture proposes that some statistical characteristics of empirically observed neuronal activity can be understood by assuming that brain networks operate in a dynamical regime near the edge of a phase transition. Moreover, the resulting critical behavior, with its concomitant scale invariance, is assumed to carry crucial functional advantages. Here, we present a data-driven analysis based on simultaneous high-throughput recordings of the activity of thousands of individual neurons in various regions of the mouse brain. To analyze these data, we construct a unified theoretical framework that synergistically combines cutting-edge methods for the study of brain activity (such as a phenomenological renormalization group approach and techniques that infer the general dynamical state of a neural population), while designing complementary tools. This unified approach allows us to uncover strong signatures of scale invariance that is "quasi-universal" across brain regions and reveal that these areas operate, to a greater or lesser extent, at the edge of instability. Furthermore, this framework allows us to distinguish between quasi-universal background activity and non-universal input-related activity. Taken together, the following study provides strong evidence that brain networks actually operate in a critical regime which, among other functional advantages, provides them with a scale-invariant substrate of activity in which optimal input representations can be sustained.

A deep learning framework for inference of single-trial neural population activity from calcium imaging with sub-frame temporal resolution

In many brain areas, neural populations act as a coordinated network whose state is tied to behavior on a moment-by-moment basis and millisecond timescale. Two-photon (2p) calcium imaging is a powerful tool to probe network-scale computation, as it can measure the activity of many individual neurons, monitor multiple layers simultaneously, and sample from identified cell types. However, estimating network states and dynamics from 2p measurements has proven challenging because of noise, inherent nonlinearities, and limitations on temporal resolution. Here we describe RADICaL, a deep learning method to overcome these limitations at the population level. RADICaL extends methods that exploit dynamics in spiking activity for application to deconvolved calcium signals, whose statistics and temporal dynamics are quite distinct from electrophysiologically-recorded spikes. It incorporates a novel network training strategy that exploits the timing of 2p sampling to recover network dynamics with high temporal precision. In synthetic tests, RADICaL infers network states more accurately than previous methods, particularly for high-frequency components. In real 2p recordings from sensorimotor areas in mice performing a "water grab" task, RADICaL infers network states with close correspondence to single-trial variations in behavior, and maintains high-quality inference even when neuronal populations are substantially reduced.

Intrinsic bursts facilitate learning of Lévy flight movements in recurrent neural network models

Isolated spikes and bursts of spikes are thought to provide the two major modes of information coding by neurons. Bursts are known to be crucial for fundamental processes between neuron pairs, such as neuronal communications and synaptic plasticity. Deficits in neuronal bursting can also impair higher cognitive functions and cause mental disorders. Despite these findings on the roles of bursts, whether and how bursts have an advantage over isolated spikes in the network-level computation remains elusive. Here, we demonstrate in a computational model that not isolated spikes but intrinsic bursts can greatly facilitate learning of Lévy flight random walk trajectories by synchronizing burst onsets across neural population. Lévy flight is a hallmark of optimal search strategies and appears in cognitive behaviors such as saccadic eye movements and memory retrieval. Our results suggest that bursting is a crucial component of sequence learning by recurrent neural networks in the brain.