Search

Grid cells in the medial entorhinal cortex (MEC) encode position using a distributed representation across multiple neural populations (modules), each possessing a distinct spatial scale. The modular structure of the representation confers the grid cell neural code with large capacity. Yet, the modularity poses significant challenges for the neural circuitry that maintains the representation, and updates it based on self motion. Small incompatible drifts in different modules, driven by noise, can rapidly lead to large, abrupt shifts in the represented position, resulting in catastrophic readout errors. Here, we propose a theoretical model of coupled modules. The coupling suppresses incompatible drifts, allowing for a stable embedding of a two-dimensional variable (position) in a higher dimensional neural attractor, while preserving the large capacity. We propose that coupling of this type may be implemented by recurrent synaptic connectivity within the MEC with a relatively simple and biologically plausible structure.

The detection of gravito-inertial forces by the otolith system is essential for our sense of balance and accurate perception. To date, however, how this system encodes the self-motion stimuli that are experienced during everyday activities remains unknown. Here, we addressed this fundamental question directly by recording from single otolith afferents in monkeys during naturalistic translational self-motion and changes in static head orientation. Otolith afferents with higher intrinsic variability transmitted more information overall about translational self-motion than their regular counterparts, owing to stronger nonlinearities that enabled precise spike timing including phase locking. By contrast, more regular afferents better discriminated between different static head orientations relative to gravity. Using computational methods, we further demonstrated that coupled increases in intrinsic variability and sensitivity accounted for the observed functional differences between afferent classes. Together, our results indicate that irregular and regular otolith afferents use different strategies to encode naturalistic self-motion and static head orientation relative to gravity.

Excitation-inhibition (EI) balance controls excitability, dynamic range, and input gating in many brain circuits. Subsets of synaptic input can be selected or 'gated' by precise modulation of finely tuned EI balance, but assessing the granularity of EI balance requires combinatorial analysis of excitatory and inhibitory inputs. Using patterned optogenetic stimulation of mouse hippocampal CA3 neurons, we show that hundreds of unique CA3 input combinations recruit excitation and inhibition with a nearly identical ratio, demonstrating precise EI balance at the hippocampus. Crucially, the delay between excitation and inhibition decreases as excitatory input increases from a few synapses to tens of synapses. This creates a dynamic millisecond-range window for postsynaptic excitation, controlling membrane depolarization amplitude and timing via subthreshold divisive normalization. We suggest that this combination of precise EI balance and dynamic EI delays forms a general mechanism for millisecond-range input gating and subthreshold gain control in feedforward networks.

A key problem in systems neuroscience is to characterize how populations of neurons encode information in their patterns of activity. An understanding of the encoding process is essential both for gaining insight into the origins of perception and for the development of brain-computer interfaces. However, this characterization is complicated by the highly variable nature of neural responses, and thus usually requires probabilistic methods for analysis. Drawing on techniques from statistical modeling and machine learning, we review recent methods for extracting important variables that quantitatively describe how sensory information is encoded in neural activity. In particular, we discuss methods for estimating receptive fields, modeling neural population dynamics, and inferring low dimensional latent structure from a population of neurons, in the context of both electrophysiology and calcium imaging data.

Artificial Neural Networks (ANNs) are bio-inspired models of neural computation that have proven highly effective. Still, ANNs lack a natural notion of time, and neural units in ANNs exchange analog values in a frame-based manner, a computationally and energetically inefficient form of communication. This contrasts sharply with biological neurons that communicate sparingly and efficiently using isomorphic binary spikes. While Spiking Neural Networks (SNNs) can be constructed by replacing the units of an ANN with spiking neurons (Cao et al., 2015; Diehl et al., 2015) to obtain reasonable performance, these SNNs use Poisson spiking mechanisms with exceedingly high firing rates compared to their biological counterparts. Here we show how spiking neurons that employ a form of neural coding can be used to construct SNNs that match high-performance ANNs and match or exceed state-of-the-art in SNNs on important benchmarks, while requiring firing rates compatible with biological findings. For this, we use spike-based coding based on the firing rate limiting adaptation phenomenon observed in biological spiking neurons. This phenomenon can be captured in fast adapting spiking neuron models, for which we derive the effective transfer function. Neural units in ANNs trained with this transfer function can be substituted directly with adaptive spiking neurons, and the resulting Adaptive SNNs (AdSNNs) can carry out competitive classification in deep neural networks without further modifications. Adaptive spike-based coding additionally allows for the dynamic control of neural coding precision: we show empirically how a simple model of arousal in AdSNNs further halves the average required firing rate and this notion naturally extends to other forms of attention as studied in neuroscience. AdSNNs thus hold promise as a novel and sparsely active model for neural computation that naturally fits to temporally continuous and asynchronous applications.

Brain signal patterns generated in the central nervous system of brain-computer interface (BCI) users are closely related to BCI paradigms and neural coding. In BCI systems, BCI paradigms and neural coding are critical elements for BCI research. However, so far there have been few references that clearly and systematically elaborated on the definition and design principles of the BCI paradigm as well as the definition and modeling principles of BCI neural coding. Therefore, these contents are expounded and the existing main BCI paradigms and neural coding are introduced in the review. Finally, the challenges and future research directions of BCI paradigm and neural coding were discussed, including user-centered design and evaluation for BCI paradigms and neural coding, revolutionizing the traditional BCI paradigms, breaking through the existing techniques for collecting brain signals and combining BCI technology with advanced AI technology to improve brain signal decoding performance. It is expected that the review will inspire innovative research and development of the BCI paradigm and neural coding., Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest., (Copyright © 2024 Tai, Ding, Wang, Gong, Li, Zhao, Su and Fu.)

It is commonly assumed that the brain’s neural coding strategies are adapted to the statistics of natural stimuli. Specifically, to maximize information transmission, a sensory neuron’s tuning function should effectively oppose the decaying stimulus spectral power, such that the neural response is temporally decorrelated (i.e. ‘whitened’). However, theory predicts that the structure of neuronal variability also plays an essential role in determining how coding is optimized. Here, we provide experimental evidence supporting this view by recording from neurons in early vestibular pathways during naturalistic self-motion. We found that central vestibular neurons displayed temporally whitened responses that could not be explained by their tuning alone. Rather, computational modeling and analysis revealed that neuronal variability and tuning were matched to effectively complement natural stimulus statistics, thereby achieving temporal decorrelation and optimizing information transmission. Taken together, our findings reveal a novel strategy by which neural variability contributes to optimized processing of naturalistic stimuli.

The nature of neural codes is central to neuroscience. Do neurons encode information through relatively slow changes in the firing rates of individual spikes (rate code) or by the precise timing of every spike (temporal code)? Here we compare the loss of information due to correlations for these two possible neural codes. The essence of Shannon's definition of information is to combine information with uncertainty: the higher the uncertainty of a given event, the more information is conveyed by that event. Correlations can reduce uncertainty or the amount of information, but by how much? In this paper we address this question by a direct comparison of the information per symbol conveyed by the words coming from a binary Markov source (temporal code) with the information per symbol coming from the corresponding Bernoulli source (uncorrelated, rate code). In a previous paper we found that a crucial role in the relation between information transmission rates (ITRs) and firing rates is played by a parameter s , which is the sum of transition probabilities from the no-spike state to the spike state and vice versa. We found that in this case too a crucial role is played by the same parameter s. We calculated the maximal and minimal bounds of the quotient of ITRs for these sources. Next, making use of the entropy grouping axiom, we determined the loss of information in a Markov source compared with the information in the corresponding Bernoulli source for a given word length. Our results show that in the case of correlated signals the loss of information is relatively small, and thus temporal codes, which are more energetically efficient, can replace rate codes effectively. These results were confirmed by experiments. [ABSTRACT FROM AUTHOR]

Environmental signals act as input and are processed across successive stages in the brain to generate a meaningful behavioral output. However, a ubiquitous observation is that descending feedback projections from more central to more peripheral brain areas vastly outnumber ascending feedforward projections. Such projections generally act to modify how sensory neurons respond to afferent signals. Recent studies in the electrosensory system of weakly electric fish have revealed novel functions for feedback pathways in that their transformation of the afferent input generates neural firing rate responses to sensory signals mediating perception and behavior. In this review, we focus on summarizing these novel and recently uncovered functions and put them into context by describing the more "classical" functions of feedback in the electrosensory system. We further highlight the parallels between the electrosensory system and other systems as well as outline interesting future directions. [ABSTRACT FROM AUTHOR]

Goal-driven and feedforward-only convolutional neural networks (CNN) have been shown to be able to predict and decode cortical responses to natural images or videos. Here, we explored an alternative deep neural network, variational auto-encoder (VAE), as a computational model of the visual cortex. We trained a VAE with a five-layer encoder and a five-layer decoder to learn visual representations from a diverse set of unlabeled images. Using the trained VAE, we predicted and decoded cortical activity observed with functional magnetic resonance imaging (fMRI) from three human subjects passively watching natural videos. Compared to CNN, VAE could predict the video-evoked cortical responses with comparable accuracy in early visual areas, but relatively lower accuracy in higher-order visual areas. The distinction between CNN and VAE in terms of encoding performance was primarily attributed to their different learning objectives, rather than their different model architecture or number of parameters. Despite lower encoding accuracies, VAE offered a more convenient strategy for decoding the fMRI activity to reconstruct the video input, by first converting the fMRI activity to the VAE's latent variables, and then converting the latent variables to the reconstructed video frames through the VAE's decoder. This strategy was more advantageous than alternative decoding methods, e.g. partial least squares regression, for being able to reconstruct both the spatial structure and color of the visual input. Such findings highlight VAE as an unsupervised model for learning visual representation, as well as its potential and limitations for explaining cortical responses and reconstructing naturalistic and diverse visual experiences. • Variational auto-encoder implements 1 an unsupervised model of "Bayesian brain". • Variational auto-encoder explains and predicts fMRI responses to natural videos. • Variational auto-encoder decodes fMRI responses to directly reconstruct visual input. • Convolutional neural networks trained for image classification better predict fMRI responses than variational auto-encoder trained for image reconstruction. [ABSTRACT FROM AUTHOR]

Abstract A convex code is a binary code generated by the pattern of intersections of a collection of open convex sets in some Euclidean space. Convex codes are relevant to neuroscience as they arise from the activity of neurons that have convex receptive fields. In this paper, we develop algebraic methods to determine if a code is convex. Specifically, we use the neural ideal of a code, which is a generalization of the Stanley–Reisner ideal. Using the neural ideal together with its standard generating set, the canonical form , we provide algebraic signatures of certain families of codes that are non-convex. We connect these signatures to the precise conditions on the arrangement of sets that prevent the codes from being convex. Finally, we also provide algebraic signatures for some families of codes that are convex, including the class of intersection-complete codes. These results allow us to detect convexity and non-convexity in a variety of situations, and point to some interesting open questions. [ABSTRACT FROM AUTHOR]

To understand how anatomy and physiology allow an organism to perform its function, it is important to know how information that is transmitted by spikes in the brain is received and encoded. A natural question is whether the spike rate alone encodes the information about a stimulus (rate code), or additional information is contained in the temporal pattern of the spikes (temporal code). Here we address this question using data from the cat Lateral Geniculate Nucleus (LGN), which is the visual portion of the thalamus, through which visual information from the retina is communicated to the visual cortex. We analyzed the responses of LGN neurons to spatially homogeneous spots of various sizes with temporally random luminance modulation. We compared the Firing Rate with the Shannon Information Transmission Rate , which quantifies the information contained in the temporal relationships between spikes. We found that the behavior of these two rates can differ quantitatively. This suggests that the energy used for spiking does not translate directly into the information to be transmitted. We also compared Firing Rates with Information Rates for X-ON and X-OFF cells. We found that, for X-ON cells the Firing Rate and Information Rate often behave in a completely different way, while for X-OFF cells these rates are much more highly correlated. Our results suggest that for X-ON cells a more efficient "temporal code" is employed, while for X-OFF cells a straightforward "rate code" is used, which is more reliable and is correlated with energy consumption. [ABSTRACT FROM AUTHOR]

Neurons in sensory areas of the neocortex are known to represent information both about sensory stimuli and behavioral state, but how these 2 disparate signals are integrated across cortical layers is poorly understood. To study this issue, we measured the coding of visual stimulus orientation and of behavioral state by neurons within superficial and deep layers of area V4 in monkeys while they covertly attended or prepared eye movements to visual stimuli. We show that whereas single neurons and neuronal populations in the superficial layers conveyed more information about the orientation of visual stimuli than neurons in deep layers, the opposite was true of information about the behavioral relevance of those stimuli. In particular, deep layer neurons encoded greater information about the direction of planned eye movements than superficial neurons. These results suggest a division of labor between cortical layers in the coding of visual input and visually guided behavior. [ABSTRACT FROM AUTHOR]

Multiplexing refers to the simultaneous encoding of two or more signals. Neurons have been shown tomultiplex, but different stimuli require different multiplexing strategies. Whereas the frequency and amplitude of periodic stimuli can be encoded by the timing and rate of the same spikes, natural scenes, which comprise areas over which intensity varies gradually and sparse edges where intensity changes abruptly, require a different multiplexing strategy. Recording in vivo from neurons in primary somatosensory cortex during tactile stimulation, we found that stimulus onset and offset (edges) evoked highly synchronized spiking, whereas other spikes in the same neurons occurred asynchronously. Stimulus intensity modulated the rate of asynchronous spiking, but did not affect the timing of synchronous spikes. From this, we hypothesized that spikes driven by high- and low-contrast stimulus features can be distinguished on the basis of their synchronization, and that differentially synchronized spiking can thus be used to form multiplexed representations. Applying a Bayesian decoding method, we verified that information about high- and low-contrast features can be recovered from an ensemble of model neurons receiving common input. Equally good decoding was achieved by distinguishing synchronous from asynchronous spikes and applying reverse correlation methods separately to each spike type. This result, which we verified with patch clamp recordings in vitro, demonstrates that neurons receiving common input can use the rate of asynchronous spiking to encode the intensity of low-contrast features while using the timing of synchronous spikes to encode the occurrence of high-contrast features. We refer to this strategy as synchrony-division multiplexing. [ABSTRACT FROM AUTHOR]

All an animal can do to infer the state of its environment is to observe the sensory-evoked activity of its own neurons. These inferences about the presence, quality, or similarity of objects are probabilistic and inform behavioral decisions that are often made in close to real time. Neural systems employ several strategies to facilitate sensory discrimination: Biophysical mechanisms separate the neuronal response distributions in coding space, compress their variances, and combine information from sequential observations. We review how these strategies are implemented in the olfactory system of the fruit fly. The emerging principles of odor discrimination likely apply to other neural circuits of similar architecture. [ABSTRACT FROM AUTHOR]

Theories of neural coding in the taste system typically rely exclusively on data gleaned from taste-responsive cells. However, even in the nucleus tractus solitarius (NTS), the first stage of central processing, neurons with taste selectivity coexist with neurons whose activity is linked to motor behavior related to ingestion. We recorded from a large ( n = 324) sample of NTS neurons recorded in awake rats, examining both their taste selectivity and the association of their activity with licking. All subjects were implanted with a bundle of microelectrodes aimed at the NTS and allowed to recover. Following moderate water deprivation, rats were placed in an experimental chamber where tastants or artificial saliva (AS) were delivered from a lick spout. Electrophysiological responses were recorded, and waveforms from single cells were isolated offline. Results showed that only a minority of NTS cells responded to taste stimuli as determined by conventional firing-rate measures. In contrast, most cells, including taste-responsive cells, tracked the lick pattern, as evidenced by significant lick coherence in the 5- to 7-Hz range. Several quantitative measures of taste selectivity and lick relatedness showed that the population formed a continuum, ranging from cells dominated by taste responses to those dominated by lick relatedness. Moreover, even neurons whose responses were highly correlated with lick activity could convey substantial information about taste quality. In all, data point to a blurred boundary between taste-dominated and lick-related cells in NTS, suggesting that information from the taste of food and from the movements it evokes are seamlessly integrated. NEW & NOTEWORTHY Neurons in the rostral nucleus of the solitary tract (NTS) are known to encode information about taste. However, recordings from awake rats reveal that only a minority of NTS cells respond exclusively to taste stimuli. The majority of neurons track behaviors associated with food consumption, and even strongly lick-related neurons could convey information about taste quality. These findings suggest that the NTS integrates information from both taste and behavior to identify food. [ABSTRACT FROM AUTHOR]

A key problem in systems neuroscience is to characterize how populations of neurons encode information in their patterns of activity. An understanding of the encoding process is essential both for gaining insight into the origins of perception and for the development of brain-computer interfaces. However, this characterization is complicated by the highly variable nature of neural responses, and thus usually requires probabilistic methods for analysis. Drawing on techniques from statistical modeling and machine learning, we review recent methods for extracting important variables that quantitatively describe how sensory information is encoded in neural activity. In particular, we discuss methods for estimating receptive fields, modeling neural population dynamics, and inferring low dimensional latent structure from a population of neurons, in the context of both electrophysiology and calcium imaging data. [ABSTRACT FROM AUTHOR]

Multiple parallel neural pathways link sound-related signals to behavioral responses. For instance, the striatum, a brain structure involved in action selection and reward-related learning, receives neuronal projections from both the auditory thalamus and auditory cortex. It is not clear whether sound information that reaches the striatum through these two pathways is redundant or complementary. We used an optogenetic approach in awake mice of both sexes to identify thalamostriatal and corticostriatal neurons during extracellular recordings, and characterized neural responses evoked by sounds of different frequencies and amplitude modulation rates. We found that neurons in both pathways encode sound frequency with similar fidelity, but display different coding strategies for amplitude modulated noise. Whereas corticostriatal neurons provide a more accurate representation of amplitude modulation rate in their overall firing rate, thalamostriatal neurons convey information about the precise timing of acoustic events. These results demonstrate that auditory thalamus and auditory cortex neurons provide complementary information to the striatum, and suggest that these pathways could be differentially recruited depending on the requirements of a sound-driven behavior. [ABSTRACT FROM AUTHOR]

Information transfer in the brain relies upon energetically expensive spiking activity of neurons. Rates of information flow should therefore be carefully optimized, but mechanisms to control this parameter are poorly understood. We address this deficit in the visual system, where ambient light (irradiance) is predictive of the amount of information reaching the eye and ask whether a neural measure of irradiance can therefore be used to proactively control information flow along the optic nerve. We first show that firing rates for the retina's output neurons [retinal ganglion cells (RGCs)] scale with irradiance and are positively correlated with rates of information and the gain of visual responses. Irradiance modulates firing in the absence of any other visual signal confirming that this is a genuine response to changing ambient light. Irradiance-driven changes in firing are observed across the population of RGCs (including in both ON and OFF units) but are disrupted in mice lacking melanopsin [the photopigment of irradiance-coding intrinsically photosensitive RGCs (ipRGCs)] and can be induced under steady light exposure by chemogenetic activation of ipRGCs. Artificially elevating firing by chemogenetic excitation of ipRGCs is sufficient to increase information flow by increasing the gain of visual responses, indicating that enhanced firing is a cause of increased information transfer at higher irradiance. Our results establish a retinal circuitry driving changes in RGC firing as an active response to alterations in ambient light to adjust the amount of visual information transmitted to the brain. [ABSTRACT FROM AUTHOR]

Deep Learning offers flexible powerful tools that have advanced our understanding of the neural coding of neurosensory systems. In this work, a 3D Convolutional Neural Network (3D CNN) is used to mimic the behavior of a population of mice retinal ganglion cells in response to different light patterns. For this purpose, we projected homogeneous RGB flashes and checkerboards stimuli with variable luminances and wavelength spectrum to mimic a more naturalistic stimuli environment onto the mouse retina. We also used white moving bars in order to localize the spatial position of the recorded cells. Then recorded spikes were smoothed with a Gaussian kernel and used as the output target when training a 3D CNN in a supervised way. To find a suitable model, two hyperparameter search stages were performed. In the first stage, a trial and error process allowed us to obtain a system that is able to fit the neurons firing rates. In the second stage, a systematic procedure was used to compare several gradient-based optimizers, loss functions and the model's convolutional layers number. We found that a three layered 3D CNN was able to predict the ganglion cells firing rates with high correlations and low prediction error, as measured with Mean Squared Error and Dynamic Time Warping in test sets. These models were either competitive or outperformed other models used already in neuroscience, as Feed Forward Neural Networks and Linear-Nonlinear models. This methodology allowed us to capture the temporal dynamic response patterns in a robust way, even for neurons with high trial-to-trial variable spontaneous firing rates, when providing the peristimulus time histogram as an output to our model. [ABSTRACT FROM AUTHOR]

This review is a curated discussion of the relationship between the gustatory system and the perception of food beginning at the earliest stage of neural processing. A brief description of the idea of taste qualities and mammalian anatomy of the taste system is presented first, followed by an overview of theories of taste coding. The case is made that food is encoded by the several senses that it stimulates beginning in the brainstem and extending throughout the entire gustatory neuraxis. In addition, the feedback from food-related movements is seamlessly melded with sensory input to create the representation of food objects in the brain.

If regularity in data takes the form of higher-order functions among groups of variables, models which are biased towards lower-order functions may easily mistake the data for noise. To distinguish whether this is the case, one must be able to quantify the contribution of different orders of dependence to the total information. Recent work in information theory attempts to do this through measures of multivariate mutual information (MMI) and information decomposition (ID). Despite substantial theoretical progress, practical issues related to tractability and learnability of higher-order functions are still largely unaddressed. In this work, we introduce a new approach to information decomposition—termed Neural Information Decomposition (NID)—which is both theoretically grounded, and can be efficiently estimated in practice using neural networks. We show on synthetic data that NID can learn to distinguish higher-order functions from noise, while many unsupervised probability models cannot. Additionally, we demonstrate the usefulness of this framework as a tool for exploring biological and artificial neural networks.

Pain and itch coding mechanisms in polymodal sensory neurons remain elusive. MrgprD+ neurons represent a major polymodal population and mediate both mechanical pain and nonhistaminergic itch. Here, we show that chemogenetic activation of MrgprD+ neurons elicited both pain- and itch-related behavior in a dose-dependent manner, revealing an unanticipated compatibility between pain and itch in polymodal neurons. While VGlut2-dependent glutamate release is required for both pain and itch transmission from MrgprD+ neurons, the neuropeptide neuromedin B (NMB) is selectively required for itch signaling. Electrophysiological recordings further demonstrated that glutamate synergizes with NMB to excite NMB-sensitive postsynaptic neurons. Ablation of these spinal neurons selectively abolished itch signals from MrgprD+ neurons, without affecting pain signals, suggesting a dedicated itch-processing central circuit. These findings reveal distinct neurotransmitters and neural circuit requirements for pain and itch signaling from MrgprD+ polymodal sensory neurons, providing new insights on coding and processing of pain and itch. [Display omitted] • Polymodal MrgprD+ primary afferent neurons express Vglut2 and Nmb • Glutamate is required for both pain and itch signaling from MrgprD+ neurons • NMB is selectively required for itch signaling • NMBR+ central neurons are required for MrgprD+-neuron-mediated itch signals Guo et al. identify the pain and itch coding mechanisms of MrgprD+ primary afferent neurons. While only glutamate is required for mechanical pain signals mediated by this population, the neuropeptide neuromedin B (NMB) and a subthreshold quantity of glutamate are released together to excite NMB-sensitive postsynaptic circuits for itch signals. [ABSTRACT FROM AUTHOR]

We study how sensory neurons detect and transmit a weak external stimulus. We use the FitzHugh–Nagumo model to simulate the neuronal activity. We consider a sub-threshold stimulus, i.e., the stimulus is below the threshold needed for triggering action potentials (spikes). However, in the presence of noise the neuron that perceives the stimulus fires a sequence of action potentials (a spike train) that carries the stimulus’ information. To yield light on how the stimulus’ information can be encoded and transmitted, we consider the simplest case of two coupled neurons, such that one neuron (referred to as neuron 1) perceives a subthreshold periodic signal but the second neuron (neuron 2) does not perceive the signal. We show that, for appropriate coupling and noise strengths, both neurons fire spike trains that have symbolic patterns (defined by the temporal structure of the inter-spike intervals), whose frequencies of occurrence depend on the signal’s amplitude and period, and are similar for both neurons. In this way, the signal information encoded in the spike train of neuron 1 propagates to the spike train of neuron 2. Our results suggest that sensory neurons can exploit the presence of neural noise to fire spike trains where the information of a subthreshold stimulus is encoded in over expressed and/or in less expressed symbolic patterns.

A fundamental task of the ascending auditory system is to produce representations that facilitate the recognition of complex sounds. This is particularly challenging in the context of acoustic variability, such as that between different talkers producing the same phoneme. These representations are transformed as information is propagated throughout the ascending auditory system from the inner ear to the auditory cortex (AI). Investigating these transformations and their role in speech recognition is key to understanding hearing impairment and the development of future clinical interventions. Here, we obtained neural responses to an extensive set of natural vowel-consonant-vowel phoneme sequences, each produced by multiple talkers, in three stages of the auditory processing pathway. Auditory nerve (AN) representations were simulated using a model of the peripheral auditory system and extracellular neuronal activity was recorded in the inferior colliculus (IC) and primary auditory cortex (AI) of anaesthetized guinea pigs. A classifier was developed to examine the efficacy of these representations for recognizing the speech sounds. Individual neurons convey progressively less information from AN to AI. Nonetheless, at the population level, representations are sufficiently rich to facilitate recognition of consonants with a high degree of accuracy at all stages indicating a progression from a dense, redundant representation to a sparse, distributed one. We examined the timescale of the neural code for consonant recognition and found that optimal timescales increase throughout the ascending auditory system from a few milliseconds in the periphery to several tens of milliseconds in the cortex. Despite these longer timescales, we found little evidence to suggest that representations up to the level of AI become increasingly invariant to across-talker differences. Instead, our results support the idea that the role of the subcortical auditory system is one of dimensionality expansion, which could provide a basis for flexible classification of arbitrary speech sounds.

Growing evidence suggests that sensory neurons achieve optimal encoding by matching their tuning properties to the natural stimulus statistics. However, the underlying mechanisms remain unclear. Here we demonstrate that feedback pathways from higher brain areas mediate optimized encoding of naturalistic stimuli via temporal whitening in the weakly electric fish Apteronotus leptorhynchus. While one source of direct feedback uniformly enhances neural responses, a separate source of indirect feedback selectively attenuates responses to low frequencies, thus creating a high-pass neural tuning curve that opposes the decaying spectral power of natural stimuli. Additionally, we recorded from two populations of higher brain neurons responsible for the direct and indirect descending inputs. While one population displayed broadband tuning, the other displayed high-pass tuning and thus performed temporal whitening. Hence, our results demonstrate a novel function for descending input in optimizing neural responses to sensory input through temporal whitening that is likely to be conserved across systems and species.

Repeating spatiotemporal spike patterns exist and carry information. Here we investigated how a single spiking neuron can optimally respond to one given pattern (localist coding), or to either one of several patterns (distributed coding, i.e., the neuron's response is ambiguous but the identity of the pattern could be inferred from the response of multiple neurons), but not to random inputs. To do so, we extended a theory developed in a previous paper (Masquelier, 2017), which was limited to localist coding. More specifically, we computed analytically the signal-to-noise ratio (SNR) of a multi-pattern-detector neuron, using a threshold-free leaky integrate-and-fire (LIF) neuron model with non-plastic unitary synapses and homogeneous Poisson inputs. Surprisingly, when increasing the number of patterns, the SNR decreases slowly, and remains acceptable for several tens of independent patterns. In addition, we investigated whether spike-timing-dependent plasticity (STDP) could enable a neuron to reach the theoretical optimal SNR. To this aim, we simulated a LIF equipped with STDP, and repeatedly exposed it to multiple input spike patterns, embedded in equally dense Poisson spike trains. The LIF progressively became selective to every repeating pattern with no supervision, and stopped discharging during the Poisson spike trains. Furthermore, tuning certain STDP parameters, the resulting pattern detectors were optimal. Tens of independent patterns could be learned by a single neuron using a low adaptive threshold, in contrast with previous studies, in which higher thresholds led to localist coding only. Taken together these results suggest that coincidence detection and STDP are powerful mechanisms, fully compatible with distributed coding. Yet we acknowledge that our theory is limited to single neurons, and thus also applies to feed-forward networks, but not to recurrent ones.

Neuroscience has long focused on finding encoding models that effectively ask “what predicts neural spiking?” and generalized linear models (GLMs) are a typical approach. It is often unknown how much of explainable neural activity is captured, or missed, when fitting a model. Here we compared the predictive performance of simple models to three leading machine learning methods: feedforward neural networks, gradient boosted trees (using XGBoost), and stacked ensembles that combine the predictions of several methods. We predicted spike counts in macaque motor (M1) and somatosensory (S1) cortices from standard representations of reaching kinematics, and in rat hippocampal cells from open field location and orientation. Of these methods, XGBoost and the ensemble consistently produced more accurate spike rate predictions and were less sensitive to the preprocessing of features. These methods can thus be applied quickly to detect if feature sets relate to neural activity in a manner not captured by simpler methods. Encoding models built with a machine learning approach accurately predict spike rates and can offer meaningful benchmarks for simpler models.

Brain activity during wakefulness is characterized by rapid fluctuations in neuronal responses. Whether these fluctuations play any role in modulating the accuracy of behavioral responses is poorly understood. Here, we investigated whether and how trial changes in the population response impact sensory coding in monkey V1 and perceptual performance. Although the responses of individual neurons varied widely across trials, many cells tended to covary with the local population. When population activity was in a ‘low’ state, neurons had lower evoked responses and correlated variability, yet higher probability to predict perceptual accuracy. The impact of firing rate fluctuations on network and perceptual accuracy was strongest 200 ms before stimulus presentation, and it greatly diminished when the number of cells used to measure the state of the population was decreased. These findings indicate that enhanced perceptual discrimination occurs when population activity is in a ‘silent’ response mode in which neurons increase information extraction.