End-to-end neural system identification with neural information flow

Neural information flow (NIF) provides a novel approach for system identification in neuroscience. It models the neural computations in multiple brain regions and can be trained end-to-end via stochastic gradient descent from noninvasive data. NIF models represent neural information processing via a network of coupled tensors, each encoding the representation of the sensory input contained in a brain region. The elements of these tensors can be interpreted as cortical columns whose activity encodes the presence of a specific feature in a spatiotemporal location. Each tensor is coupled to the measured data specific to a brain region via low-rank observation models that can be decomposed into the spatial, temporal and feature receptive fields of a localized neuronal population. Both these observation models and the convolutional weights defining the information processing within regions are learned end-to-end by predicting the neural signal during sensory stimulation. We trained a NIF model on the activity of early visual areas using a large-scale fMRI dataset recorded in a single participant. We show that we can recover plausible visual representations and population receptive fields that are consistent with empirical findings.     

Analysis of Slow (Theta) Oscillations as a Potential Temporal Reference Frame for Information Coding in Sensory Cortices

While sensory neurons carry behaviorally relevant information in responses that often extend over hundreds of milliseconds, the key units of neural information likely consist of much shorter and temporally precise spike patterns. The mechanisms and temporal reference frames by which sensory networks partition responses into these shorter units of information remain unknown. One hypothesis holds that slow oscillations provide a network-intrinsic reference to temporally partitioned spike trains without exploiting the millisecond-precise alignment of spikes to sensory stimuli. We tested this hypothesis on neural responses recorded in visual and auditory cortices of macaque monkeys in response to natural stimuli. Comparing different schemes for response partitioning revealed that theta band oscillations provide a temporal reference that permits extracting significantly more information than can be obtained from spike counts, and sometimes almost as much information as obtained by partitioning spike trains using precisely stimulus-locked time bins. We further tested the robustness of these partitioning schemes to temporal uncertainty in the decoding process and to noise in the sensory input. This revealed that partitioning using an oscillatory reference provides greater robustness than partitioning using precisely stimulus-locked time bins. Overall, these results provide a computational proof of concept for the hypothesis that slow rhythmic network activity may serve as internal reference frame for information coding in sensory cortices and they foster the notion that slow oscillations serve as key elements for the computations underlying perception.     

Spike suppression in a local cortical circuit induced by transcranial magnetic stimulation

Transcranial magnetic stimulation (TMS) noninvasively interferes with human cortical function, and is widely used as an effective technique for probing causal links between neural activity and cognitive function. However, the physiological mechanisms underlying TMS-induced effects on neural activity remain unclear. We examined the mechanism by which TMS disrupts neural activity in a local circuit in early visual cortex using a computational model consisting of conductance-based spiking neurons with excitatory and inhibitory synaptic connections. We found that single-pulse TMS suppressed spiking activity in a local circuit model, disrupting the population response. Spike suppression was observed when TMS was applied to the local circuit within a limited time window after the local circuit received sensory afferent input, as observed in experiments investigating suppression of visual perception with TMS targeting early visual cortex. Quantitative analyses revealed that the magnitude of suppression was significantly larger for synaptically-connected neurons than for isolated individual neurons, suggesting that intracortical inhibitory synaptic coupling also plays an important role in TMS-induced suppression. A conventional local circuit model of early visual cortex explained only the early period of visual suppression observed in experiments. However, models either involving strong recurrent excitatory synaptic connections or sustained excitatory input were able to reproduce the late period of visual suppression. These results suggest that TMS targeting early visual cortex disrupts functionally distinct neural signals, possibly corresponding to feedforward and recurrent information processing, by imposing inhibitory effects through intracortical inhibitory synaptic connections.     

Neural dynamics as sampling: a model for stochastic computation in recurrent networks of spiking neurons

The organization of computations in networks of spiking neurons in the brain is still largely unknown, in particular in view of the inherently stochastic features of their firing activity and the experimentally observed trial-to-trial variability of neural systems in the brain. In principle there exists a powerful computational framework for stochastic computations, probabilistic inference by sampling, which can explain a large number of macroscopic experimental data in neuroscience and cognitive science. But it has turned out to be surprisingly difficult to create a link between these abstract models for stochastic computations and more detailed models of the dynamics of networks of spiking neurons. Here we create such a link and show that under some conditions the stochastic firing activity of networks of spiking neurons can be interpreted as probabilistic inference via Markov chain Monte Carlo (MCMC) sampling. Since common methods for MCMC sampling in distributed systems, such as Gibbs sampling, are inconsistent with the dynamics of spiking neurons, we introduce a different approach based on non-reversible Markov chains that is able to reflect inherent temporal processes of spiking neuronal activity through a suitable choice of random variables. We propose a neural network model and show by a rigorous theoretical analysis that its neural activity implements MCMC sampling of a given distribution, both for the case of discrete and continuous time. This provides a step towards closing the gap between abstract functional models of cortical computation and more detailed models of networks of spiking neurons.     

Using hardware models to quantify sensory data acquisition across the rat vibrissal array

Our laboratory investigates how animals acquire sensory data to understand the neural computations that permit complex sensorimotor behaviors. We use the rat whisker system as a model to study active tactile sensing; our aim is to quantitatively describe the spatiotemporal structure of incoming sensory information to place constraints on subsequent neural encoding and processing. In the first part of this paper we describe the steps in the development of a hardware model (a 'sensobot') of the rat whisker array that can perform object feature extraction. We show how this model provides insights into the neurophysiology and behavior of the real animal. In the second part of this paper, we suggest that sensory data acquisition across the whisker array can be quantified using the complete derivative. We use the example of wall-following behavior to illustrate that computing the appropriate spatial gradients across a sensor array would enable an animal or mobile robot to predict the sensory data that will be acquired at the next time step.     

Charles Bonnet Syndrome: Evidence for a Generative Model in the Cortex?

Several theories propose that the cortex implements an internal model to explain, predict, and learn about sensory data, but the nature of this model is unclear. One condition that could be highly informative here is Charles Bonnet syndrome (CBS), where loss of vision leads to complex, vivid visual hallucinations of objects, people, and whole scenes. CBS could be taken as indication that there is a generative model in the brain, specifically one that can synthesise rich, consistent visual representations even in the absence of actual visual input. The processes that lead to CBS are poorly understood. Here, we argue that a model recently introduced in machine learning, the deep Boltzmann machine (DBM), could capture the relevant aspects of (hypothetical) generative processing in the cortex. The DBM carries both the semantics of a probabilistic generative model and of a neural network. The latter allows us to model a concrete neural mechanism that could underlie CBS, namely, homeostatic regulation of neuronal activity. We show that homeostatic plasticity could serve to make the learnt internal model robust against e.g. degradation of sensory input, but overcompensate in the case of CBS, leading to hallucinations. We demonstrate how a wide range of features of CBS can be explained in the model and suggest a potential role for the neuromodulator acetylcholine. This work constitutes the first concrete computational model of CBS and the first application of the DBM as a model in computational neuroscience. Our results lend further credence to the hypothesis of a generative model in the brain.     

Frequency separation by an excitatory-inhibitory network

We consider a situation in which individual features of the input are represented in the neural system by different frequencies of periodic firings. Thus, if two of the features are presented concurrently, the input to the system will consist of a superposition of two periodic trains. In this paper we present an algorithm that is capable of extracting the individual features from the composite signal by separating the signal into periodic spike trains with different frequencies. We show that the algorithm can be implemented in a biophysically based excitatory-inhibitory network model. The frequency separation process works over a range of frequencies determined by time constants of the model’s intrinsic variables. It does not rely on a “resonance” phenomenon and is not tuned to a discrete set of frequencies. The frequency separation is still reliable when the timing of incoming spikes is noisy.     

Functional correlates of the lateral and medial entorhinal cortex: objects,path integration and local–global reference frames

The hippocampus receives its major cortical input from the medial entorhinal cortex (MEC) and the lateral entorhinal cortex (LEC). It is commonly believed that the MEC provides spatial input to the hippocampus, whereas the LEC provides non-spatial input. We review new data which suggest that this simple dichotomy between ‘where’ versus ‘what’ needs revision. We propose a refinement of this model, which is more complex than the simple spatial–non-spatial dichotomy. MEC is proposed to be involved in path integration computations based on a global frame of reference, primarily using internally generated, self-motion cues and external input about environmental boundaries and scenes; it provides the hippocampus with a coordinate system that underlies the spatial context of an experience. LEC is proposed to process information about individual items and locations based on a local frame of reference, primarily using external sensory input; it provides the hippocampus with information about the content of an experience.     

Understanding auditory spectro-temporal receptive fields and their changes with input statistics by efficient coding principles

Spectro-temporal receptive fields (STRFs) have been widely used as linear approximations to the signal transform from sound spectrograms to neural responses along the auditory pathway. Their dependence on statistical attributes of the stimuli, such as sound intensity, is usually explained by nonlinear mechanisms and models. Here, we apply an efficient coding principle which has been successfully used to understand receptive fields in early stages of visual processing, in order to provide a computational understanding of the STRFs. According to this principle, STRFs result from an optimal tradeoff between maximizing the sensory information the brain receives, and minimizing the cost of the neural activities required to represent and transmit this information. Both terms depend on the statistical properties of the sensory inputs and the noise that corrupts them. The STRFs should therefore depend on the input power spectrum and the signal-to-noise ratio, which is assumed to increase with input intensity. We analytically derive the optimal STRFs when signal and noise are approximated as Gaussians. Under the constraint that they should be spectro-temporally local, the STRFs are predicted to adapt from being band-pass to low-pass filters as the input intensity reduces, or the input correlation becomes longer range in sound frequency or time. These predictions qualitatively match physiological observations. Our prediction as to how the STRFs should be determined by the input power spectrum could readily be tested, since this spectrum depends on the stimulus ensemble. The potentials and limitations of the efficient coding principle are discussed.     

A Lightweight,Headphones-based System for Manipulating Auditory Feedback in Songbirds

Experimental manipulations of sensory feedback during complex behavior have provided valuable insights into the computations underlying motor control and sensorimotor plasticity¹. Consistent sensory perturbations result in compensatory changes in motor output, reflecting changes in feedforward motor control that reduce the experienced feedback error. By quantifying how different sensory feedback errors affect human behavior, prior studies have explored how visual signals are used to recalibrate arm movements^2,3 and auditory feedback is used to modify speech production^4-7. The strength of this approach rests on the ability to mimic naturalistic errors in behavior, allowing the experimenter to observe how experienced errors in production are used to recalibrate motor output.Songbirds provide an excellent animal model for investigating the neural basis of sensorimotor control and plasticity^8,9. The songbird brain provides a well-defined circuit in which the areas necessary for song learning are spatially separated from those required for song production, and neural recording and lesion studies have made significant advances in understanding how different brain areas contribute to vocal behavior^9-12. However, the lack of a naturalistic error-correction paradigm - in which a known acoustic parameter is perturbed by the experimenter and then corrected by the songbird - has made it difficult to understand the computations underlying vocal learning or how different elements of the neural circuit contribute to the correction of vocal errors¹³.The technique described here gives the experimenter precise control over auditory feedback errors in singing birds, allowing the introduction of arbitrary sensory errors that can be used to drive vocal learning. Online sound-processing equipment is used to introduce a known perturbation to the acoustics of song, and a miniaturized headphones apparatus is used to replace a songbird''s natural auditory feedback with the perturbed signal in real time. We have used this paradigm to perturb the fundamental frequency (pitch) of auditory feedback in adult songbirds, providing the first demonstration that adult birds maintain vocal performance using error correction¹⁴. The present protocol can be used to implement a wide range of sensory feedback perturbations (including but not limited to pitch shifts) to investigate the computational and neurophysiological basis of vocal learning.     

Population approach to a neural discrimination task

This article gives insights into the possible neuronal processes involved in visual discrimination. We study the performance of a spiking network of Integrate-and-Fire (IF) neurons when performing a benchmark discrimination task. The task we adopted consists of determining the direction of moving dots in a noisy context using similar stimuli to those in the experiments of Newsome and colleagues. We present a neural model that performs the discrimination involved in this task. By varying the synaptic parameters of the IF neurons, we illustrate the counter-intuitive importance of the second-order statistics (input noise) in improving the discrimination accuracy of the model. We show that measuring the Firing Rate (FR) over a population enables the model to discriminate in realistic times, and even surprisingly significantly increases its discrimination accuracy over the single neuron case, despite the faster processing. We also show that increasing the input noise increases the discrimination accuracy but only at the expense of the speed at which we can read out the FR.     

A case for spiking neural network simulation based on configurable multiple-FPGA systems

Recent neuropsychological research has begun to reveal that neurons encode information in the timing of spikes. Spiking neural network simulations are a flexible and powerful method for investigating the behaviour of neuronal systems. Simulation of the spiking neural networks in software is unable to rapidly generate output spikes in large-scale of neural network. An alternative approach, hardware implementation of such system, provides the possibility to generate independent spikes precisely and simultaneously output spike waves in real time, under the premise that spiking neural network can take full advantage of hardware inherent parallelism. We introduce a configurable FPGA-oriented hardware platform for spiking neural network simulation in this work. We aim to use this platform to combine the speed of dedicated hardware with the programmability of software so that it might allow neuroscientists to put together sophisticated computation experiments of their own model. A feed-forward hierarchy network is developed as a case study to describe the operation of biological neural systems (such as orientation selectivity of visual cortex) and computational models of such systems. This model demonstrates how a feed-forward neural network constructs the circuitry required for orientation selectivity and provides platform for reaching a deeper understanding of the primate visual system. In the future, larger scale models based on this framework can be used to replicate the actual architecture in visual cortex, leading to more detailed predictions and insights into visual perception phenomenon.     

Spiking networks for Bayesian inference and choice

Systems neuroscience traditionally conceptualizes a population of spiking neurons as merely encoding the value of a stimulus. Yet, psychophysics has revealed that people take into account stimulus uncertainty when performing sensory or motor computations and do so in a nearly Bayes-optimal way. This suggests that neural populations do not encode just a single value but an entire probability distribution over the stimulus. Several such probabilistic codes have been proposed, including one that utilizes the structure of neural variability to enable simple neural implementations of probabilistic computations such as optimal cue integration. This approach provides a quantitative link between Bayes-optimal behaviors and specific neural operations. It allows for novel ways to evaluate probabilistic codes and for predictions for physiological population recordings.     

Stimulus-dependent Maximum Entropy Models of Neural Population Codes

Neural populations encode information about their stimulus in a collective fashion, by joint activity patterns of spiking and silence. A full account of this mapping from stimulus to neural activity is given by the conditional probability distribution over neural codewords given the sensory input. For large populations, direct sampling of these distributions is impossible, and so we must rely on constructing appropriate models. We show here that in a population of 100 retinal ganglion cells in the salamander retina responding to temporal white-noise stimuli, dependencies between cells play an important encoding role. We introduce the stimulus-dependent maximum entropy (SDME) model—a minimal extension of the canonical linear-nonlinear model of a single neuron, to a pairwise-coupled neural population. We find that the SDME model gives a more accurate account of single cell responses and in particular significantly outperforms uncoupled models in reproducing the distributions of population codewords emitted in response to a stimulus. We show how the SDME model, in conjunction with static maximum entropy models of population vocabulary, can be used to estimate information-theoretic quantities like average surprise and information transmission in a neural population.     

Multisensory Causal Inference in the Brain

At any given moment, our brain processes multiple inputs from its different sensory modalities (vision, hearing, touch, etc.). In deciphering this array of sensory information, the brain has to solve two problems: (1) which of the inputs originate from the same object and should be integrated and (2) for the sensations originating from the same object, how best to integrate them. Recent behavioural studies suggest that the human brain solves these problems using optimal probabilistic inference, known as Bayesian causal inference. However, how and where the underlying computations are carried out in the brain have remained unknown. By combining neuroimaging-based decoding techniques and computational modelling of behavioural data, a new study now sheds light on how multisensory causal inference maps onto specific brain areas. The results suggest that the complexity of neural computations increases along the visual hierarchy and link specific components of the causal inference process with specific visual and parietal regions.     

Collective motion in animal groups from a neurobiological perspective: The adaptive benefits of dynamic sensory loads and selective attention

We explore mechanisms associated with collective animal motion by drawing on the neurobiological bases of sensory information processing and decision-making. The model uses simplified retinal processes to translate neighbor movement patterns into information through spatial signal integration and threshold responses. The structure provides a mechanism by which individuals can vary their sets of influential neighbors, a measure of an individual's sensory load. Sensory loads are correlated with group order and density, and we discuss their adaptive values in an ecological context. The model also provides a mechanism by which group members can identify, and rapidly respond to, novel visual stimuli.     

Sensory coding and contrast invariance emerge from the control of plastic inhibition over emergent selectivity

Visual stimuli are represented by a highly efficient code in the primary visual cortex, but the development of this code is still unclear. Two distinct factors control coding efficiency: Representational efficiency, which is determined by neuronal tuning diversity, and metabolic efficiency, which is influenced by neuronal gain. How these determinants of coding efficiency are shaped during development, supported by excitatory and inhibitory plasticity, is only partially understood. We investigate a fully plastic spiking network of the primary visual cortex, building on phenomenological plasticity rules. Our results suggest that inhibitory plasticity is key to the emergence of tuning diversity and accurate input encoding. We show that inhibitory feedback (random and specific) increases the metabolic efficiency by implementing a gain control mechanism. Interestingly, this led to the spontaneous emergence of contrast-invariant tuning curves. Our findings highlight that (1) interneuron plasticity is key to the development of tuning diversity and (2) that efficient sensory representations are an emergent property of the resulting network.     

Circuit Motifs for Contrast-Adaptive Differentiation in Early Sensory Systems: The Role of Presynaptic Inhibition and Short-Term Plasticity

In natural signals, such as the luminance value across of a visual scene, abrupt changes in intensity value are often more relevant to an organism than intensity values at other positions and times. Thus to reduce redundancy, sensory systems are specialized to detect the times and amplitudes of informative abrupt changes in the input stream rather than coding the intensity values at all times. In theory, a system that responds transiently to fast changes is called a differentiator. In principle, several different neural circuit mechanisms exist that are capable of responding transiently to abrupt input changes. However, it is unclear which circuit would be best suited for early sensory systems, where the dynamic range of the natural input signals can be very wide. We here compare the properties of different simple neural circuit motifs for implementing signal differentiation. We found that a circuit motif based on presynaptic inhibition (PI) is unique in a sense that the vesicle resources in the presynaptic site can be stably maintained over a wide range of stimulus intensities, making PI a biophysically plausible mechanism to implement a differentiator with a very wide dynamical range. Moreover, by additionally considering short-term plasticity (STP), differentiation becomes contrast adaptive in the PI-circuit but not in other potential neural circuit motifs. Numerical simulations show that the behavior of the adaptive PI-circuit is consistent with experimental observations suggesting that adaptive presynaptic inhibition might be a good candidate neural mechanism to achieve differentiation in early sensory systems.     

Adaptive Spike Threshold Enables Robust and Temporally Precise Neuronal Encoding

Neural processing rests on the intracellular transformation of information as synaptic inputs are translated into action potentials. This transformation is governed by the spike threshold, which depends on the history of the membrane potential on many temporal scales. While the adaptation of the threshold after spiking activity has been addressed before both theoretically and experimentally, it has only recently been demonstrated that the subthreshold membrane state also influences the effective spike threshold. The consequences for neural computation are not well understood yet. We address this question here using neural simulations and whole cell intracellular recordings in combination with information theoretic analysis. We show that an adaptive spike threshold leads to better stimulus discrimination for tight input correlations than would be achieved otherwise, independent from whether the stimulus is encoded in the rate or pattern of action potentials. The time scales of input selectivity are jointly governed by membrane and threshold dynamics. Encoding information using adaptive thresholds further ensures robust information transmission across cortical states i.e. decoding from different states is less state dependent in the adaptive threshold case, if the decoding is performed in reference to the timing of the population response. Results from in vitro neural recordings were consistent with simulations from adaptive threshold neurons. In summary, the adaptive spike threshold reduces information loss during intracellular information transfer, improves stimulus discriminability and ensures robust decoding across membrane states in a regime of highly correlated inputs, similar to those seen in sensory nuclei during the encoding of sensory information.     

Generation of tactile maps for artificial skin

Prior research has shown that representations of retinal surfaces can be learned from the intrinsic structure of visual sensory data in neural simulations, in robots, as well as by animals. Furthermore, representations of cochlear (frequency) surfaces can be learned from auditory data in neural simulations. Advances in hardware technology have allowed the development of artificial skin for robots, realising a new sensory modality which differs in important respects from vision and audition in its sensorimotor characteristics. This provides an opportunity to further investigate ordered sensory map formation using computational tools. We show that it is possible to learn representations of non-trivial tactile surfaces, which require topologically and geometrically involved three-dimensional embeddings. Our method automatically constructs a somatotopic map corresponding to the configuration of tactile sensors on a rigid body, using only intrinsic properties of the tactile data. The additional complexities involved in processing the tactile modality require the development of a novel multi-dimensional scaling algorithm. This algorithm, ANISOMAP, extends previous methods and outperforms them, producing high-quality reconstructions of tactile surfaces in both simulation and hardware tests. In addition, the reconstruction turns out to be robust to unanticipated hardware failure.