Computational Neuroscience

It has been argued that emotion, pain and cognitive control are functionally segregated in distinct subdivisions of the cingulate cortex. However, recent observations encourage a fundamentally different view. Imaging studies demonstrate that negative affect, pain and cognitive control activate an overlapping region of the dorsal cingulate — the anterior midcingulate cortex (aMCC). Anatomical studies reveal that the aMCC constitutes a hub where information about reinforcers can be linked to motor centres responsible for expressing affect and executing goal-directed behaviour. Computational modelling and other kinds of evidence suggest that this intimacy reflects control processes that are common to all three domains. These observations compel a reconsideration of the dorsal cingulate's contribution to negative affect and pain.

I propose that synchronization affects communication between neuronal groups. Gamma-band (30-90 Hz) synchronization modulates excitation rapidly enough so it escapes the following inhibition and activates postsynaptic neurons effectively. Synchronization also ensures that a presynaptic activation pattern arrives at postsynaptic neurons in a temporally coordinated manner. At a postsynaptic neuron, multiple presynaptic groups converge, e.g. representing different stimuli. If a stimulus is selected by attention, its neuronal representation shows stronger and higher-frequency gamma-band synchronization. Thereby, the attended stimulus representation selectively entrains postsynaptic neurons. The entrainment creates sequences of short excitation and longer inhibition that are coordinated between pre- and postsynaptic groups to transmit the attended representation and shut out competing inputs. The predominantly bottom-up directed gamma-band influences are controlled by predominantly top-down directed alpha-beta band (8-20 Hz) influences. Attention itself samples stimuli at a 7-8 Hz theta rhythm. Thus, several rhythms and their interplay render neuronal communication effective, precise and selective.

Humans and monkeys easily recognize objects in scenes. This ability is known to be supported by a network of hierarchically interconnected brain areas. However, understanding neurons in higher levels of this hierarchy has long remained a major challenge in visual systems neuroscience. We use computational techniques to identify a neural network model that matches human performance on challenging object categorization tasks. Although not explicitly constrained to match neural data, this model turns out to be highly predictive of neural responses in both the V4 and inferior temporal cortex, the top two layers of the ventral visual hierarchy. In addition to yielding greatly improved models of visual cortex, these results suggest that a process of biological performance optimization directly shaped neural mechanisms.

Donoghue, and Emery N. Brown. A point process framework for relating neural spiking activity to spiking history, neural ensemble, and extrinsic covariate effects. . Multiple factors simultaneously affect the spiking activity of individual neurons. Determining the effects and relative importance of these factors is a challenging problem in neurophysiology. We propose a statistical framework based on the point process likelihood function to relate a neuron's spiking probability to three typical covariates: the neuron's own spiking history, concurrent ensemble activity, and extrinsic covariates such as stimuli or behavior. The framework uses parametric models of the conditional intensity function to define a neuron's spiking probability in terms of the covariates. The discrete time likelihood function for point processes is used to carry out model fitting and model analysis. We show that, by modeling the logarithm of the conditional intensity function as a linear combination of functions of the covariates, the discrete time point process likelihood function is readily analyzed in the generalized linear model (GLM) framework. We illustrate our approach for both GLM and non-GLM likelihood functions using simulated data and multivariate single-unit activity data simultaneously recorded from the motor cortex of a monkey performing a visuomotor pursuit-tracking task. The point process framework provides a flexible, computationally efficient approach for maximum likelihood estimation, goodness-of-fit assessment, residual analysis, model selection, and neural decoding. The framework thus allows for the formulation and analysis of point process models of neural spiking activity that readily capture the simultaneous effects of multiple covariates and enables the assessment of their relative importance.

We review different aspects of the simulation of spiking neural networks. We start by reviewing the different types of simulation strategies and algorithms that are currently implemented. We next review the precision of those simulation strategies, in particular in cases where plasticity depends on the exact timing of the spikes. We overview different simulators and simulation environments presently available (restricted to those freely available, open source and documented). For each simulation tool, its advantages and pitfalls are reviewed, with an aim to allow the reader to identify which simulator is appropriate for a given task. Finally, we provide a series of benchmark simulations of different types of networks of spiking neurons, including Hodgkin-Huxley type, integrate-and-fire models, interacting with current-based or conductance-based synapses, using clock-driven or event-driven integration strategies. The same set of models are implemented on the different simulators, and the codes are made available. The ultimate goal of this review is to provide a resource to facilitate identifying the appropriate integration strategy and simulation tool to use for a given modeling problem related to spiking neural networks.

Understanding causal relationships, or effective connectivity, between parts of the brain is of utmost importance because a large part of the brain's activity is thought to be internally generated and, hence, quantifying stimulus response relationships alone does not fully describe brain dynamics. Past efforts to determine effective connectivity mostly relied on model based approaches such as Granger causality or dynamic causal modeling. Transfer entropy (TE) is an alternative measure of effective connectivity based on information theory. TE does not require a model of the interaction and is inherently non-linear. We investigated the applicability of TE as a metric in a test for effective connectivity to electrophysiological data Action Editor: Aurel A. Lazar R. Vicente, M. Wibral, and M. Lindner contributed equally.

The thalamocortical projection is an integral part of the primary pathway through which information from the outside world reaches the neocortex. Given such a vital task, it is no wonder that the cortical neurons onto which the TC projection synapses (L4) have large postsynaptic potentials. Nonetheless, ~85% of synapses onto cortical neurons come not from the thalamus, but from within the cortex. This raises questions about the efficacy of this projection, which the authors deal with in this paper. The researchers initially suggest different possibilities for the causes of PSPs in the barrels of rats S1, when following thalamic input. They pose the plausibility of recurrent excitatory connections within L4, as an amplification circuit to incoming thalamic input. They also suggest that the TC synapses could simply have high-efficacy, thus invalidating the need for extra circuits. Throughout the paper, findings are presented about the amplitude of aPSPs in L4, and the very structure of the thalamocortical projection, in order to discuss these possibilities. In vitro recordings show a stronger TC–L4 synapse than other CT connections, in the order of 4 mV. The authors discredit this value, obtained through means of electrical stimulation of TC fibers. Certain dynamics of synaptic depression following persistent activation seem to be absent from in vitro recordings, and these could influence the obtained uEPSP. In figure 3C, it is suggested by intracellular recordings in L4 that the thalamus' efficacy decreases if it fires repeatedly. This coherently explains why the lack of spontaneous activity in vitro contributes to larger EPSPs. The ingenious presentation of recordings also allows us to see a difference between anesthetized and sedated conditions. While the latter has substantial spontaneous rates — 5.4 Hz —, the former condition has an average of 1 second ISI. These values are in the order of those presented in figure 3C, and are consistent with the observation that " aPSPs of TC–barrel neuron connections are smaller during sedation than during anesthesia ". However, an ISI of 1 second seems to be sufficient to fully recover from synaptic depression, as the values in 3C seemingly converge to an upper bound. It is no wonder, then, that the strongest synapses in the anesthetized condition were also in the order of 4 mV, although their average is 1.94 mV. The reduced driving force in vivo can likely account for this disparity. The values of aPSPs presented above were obtained using dual recordings in the alive brain of rodents, either sedated or anesthetized. Extracellular recordings were made on the VPM nucleus, in barreloids. The low resolution of these recordings is sufficient, given that only (easily observable) action potentials have an effect on their efferents. These are also difficult areas to access using patch pipettes. The membrane potential was simultaneously recorded inside cortical neurons, mostly within layer 4 of somatotopically aligned barrels, to explore this " strong-synapse model. " While barrages of thalamic and cortical postsynaptic potentials continuously generate large enough V m fluctuations to obscure uEPSPs of single TC–L4 synapses, the researchers computed spike-triggered averages of cortical V m during periodic stimulation of VPM, and corrected for stimulus-induced correlations by subtracting the shift-predictor. This yielded the aPSP that a single thalamic AP evokes in the cortical cell, despite all the excitatory and inhibitory inputs it is a target of. The researchers were also keen in reducing the noise in the cortical neuron by modulating the amount of thalamic neurons firing in synchrony, through the choice of a slow oscillatory movement for the whisker deflections.

Stochastic resonance is said to be observed
when increases in levels of unpredictable fluctuations—
e.g., random noise—cause an increase in a metric of the
quality of signal transmission or detection performance,
rather than a decrease. This counterintuitive effect relies
on system nonlinearities and on some parameter ranges
being ‘‘suboptimal’’. Stochastic resonance has been
observed, quantified, and described in a plethora of
physical and biological systems, including neurons. Being
a topic of widespread multidisciplinary interest, the
definition of stochastic resonance has evolved significantly
over the last decade or so, leading to a number of
debates, misunderstandings, and controversies. Perhaps
the most important debate is whether the brain has
evolved to utilize random noise in vivo, as part of the
‘‘neural code’’. Surprisingly, this debate has been for the
most part ignored by neuroscientists, despite much
indirect evidence of a positive role for noise in the brain.
We explore some of the reasons for this and argue why it
would be more surprising if the brain did not exploit
randomness provided by noise—via stochastic resonance
or otherwise—than if it did. We also challenge neuroscientists
and biologists, both computational and experimental,
to embrace a very broad definition of stochastic
resonance in terms of signal-processing ‘‘noise benefits’’,
and to devise experiments aimed at verifying that random
variability can play a functional role in the brain, nervous
system, or other areas of biology.
Introduction

Markov kinetic models were used to synthesize a complete description of synaptic transmission, including opening of voltage-dependent channels in the presynaptic terminal, release of neurotransmitter, gating of postsynaptic receptors, and activation of second-messenger systems. These kinetic schemes provide a more general framework for modeling ion channels than the Hodgkin-Huxley formalism, supporting a continuous spectrum of descriptions ranging from the very simple and computationally efficient to the highly complex and biophysically precise. Examples are given of simple kinetic schemes based on fits to experimental data that capture the essential properties of voltage-gated, synaptic and neuromodulatory currents. The Markov formalism allows the dynamics of ionic currents to be considered naturally in the larger context of biochemical signal transduction. This framework can facilitate the integration of a wide range of experimental data and promote consistent theoretical analysis of neural mechanisms from molecular interactions to network computations.

Electrical signals from the cortical surface of animals were recorded as early as 1875 (REF. 1), 50 years before the advent of electroencephalography (EEG) 2 . Subsequent work revealed that the high-frequency part (above ~500 Hz) of the recorded potentials provides information about the spiking activity of neurons located around the electrode 3 . By contrast, the part of the signal that has frequencies below ~500 Hz, the so-called 'local field potential' (LFP), was found more difficult to interpret in terms of the underlying neural activity. Although the introduction of current source density (CSD) analysis in the 1950s 4 rejuvenated the use of the LFP in the following decades 5-7 , interest decreased in the 1980s and 1990s, probably owing to the focus on new single-neuron techniques (for example, patch-clamp recordings) and on understanding the link between single-neuron activity and perception.

Machine learning and pattern recognition algorithms have in the past years developed to become a working horse in brain imaging and the computational neurosciences, as they are instrumental for mining vast amounts of neural data of ever increasing measurement precision and detecting minuscule signals from an overwhelming noise floor. They provide the means to decode and characterize task relevant brain states and to distinguish them from non-informative brain signals. While undoubtedly this machinery has helped to gain novel biological insights, it also holds the danger of potential unintentional abuse. Ideally machine learning techniques should be usable for any non-expert, however, unfortunately they are typically not. Overfitting and other pitfalls may occur and lead to spurious and nonsensical interpretation. The goal of this review is therefore to provide an accessible and clear introduction to the strengths and also the inherent dangers of machine learning usage in the neurosciences.

Computational neuroscience has produced a diversity of software for simulations of networks of spiking neurons, with both negative and positive consequences. On the one hand, each simulator uses its own programming or configuration language, leading to considerable difficulty in porting models from one simulator to another. This impedes communication between investigators and makes it harder to reproduce and build on the work of others. On the other hand, simulation results can be cross-checked between different simulators, giving greater confidence in their correctness, and each simulator has different optimizations, so the most appropriate simulator can be chosen for a given modelling task. A common programming interface to multiple simulators would reduce or eliminate the problems of simulator diversity while retaining the benefits. PyNN is such an interface, making it possible to write a simulation script once, using the Python programming language, and run it without modification on any supported simulator (currently NEURON, NEST, PCSIM, Brian and the Heidelberg VLSI neuromorphic hardware). PyNN increases the productivity of neuronal network modelling by providing high-level abstraction, by promoting code sharing and reuse, and by providing a foundation for simulator-agnostic analysis, visualization and data-management tools. PyNN increases the reliability of modelling studies by making it much easier to check results on multiple simulators. PyNN is open-source software and is available from http://neuralensemble.org/PyNN.

The development of the issue of binding as fundamental to neural dynamics has made possible recent advances in the modeling of di cult problems of perception and brain function. Among them is perceptual segmentation, invariant pattern recognition and one-shot learning. Also, longer-term conceptual developments that have led to this success are reviewed.

The hippocampal formation presents a special opportunity for realistic neural modeling since its structure, connectivity, and physiology are better understood than that of other cortical components. A review of the quantitative neuroanatomy of the rodent dentate gyrus (DG) is presented in the context of the development of a computational model of its connectivity. The DG is a three-layered folded sheet of neural tissue. This sheet is represented as a rectangle, having a surface area of 37 mm2 and a septotemporal length of 12 mm. Points, representing cell somata, are distributed in the model rectangle in a roughly uniform fashion. Synaptic connectivity is generated by assigning each presynaptic cell a spatial zone representing its axonal arbor. For each postsynaptic cell, a list of potential presynaptic cells is compiled, based on which arbor zones the given postsynaptic cell falls within. An appropriate number of presynaptic inputs are then selected at random. The principal cells of the DG, the granule cells, are represented in the model, as are non-principal cells, including basket cells, chandelier cells, mossy cells, and GABAergic peptidergic polymorphic (GPP) cells. The neurons of layer II of the entorhinal cortex are included also. The DG receives its main extrinsic input from these cells via the perforant path. The basket cells, chandelier cells, and GPP cells receive perforant path and granule cell input and exert both feedforward and feedback inhibition onto the granule cells. Mossy cells receive converging input from granule cells and send their output back primarily to distant septotemporal levels, where they contact both granule cells and non-principal cells. To permit numerical simulations, the model must be scaled down while preserving its anatomical structure. A variety of methods for doing this exist. Hippocampal allometry provides valuable clues in this regard.

Paninski, Liam, Matthew R. Fellows, Nicholas G. Hatsopoulos, and John P. Donoghue. Spatiotemporal tuning of motor cortical neurons for hand position and velocity. J Neurophysiol 91: 515-532, A pursuit-tracking task (PTT) and multielectrode recordings were used to investigate the spatiotemporal encoding of hand position and velocity in primate primary motor cortex (MI). Continuous tracking of a randomly moving visual stimulus provided a broad sample of velocity and position space, reduced statistical dependencies between kinematic variables, and minimized the nonstationarities that are found in typical "step-tracking" tasks. These statistical features permitted the application of signal-processing and information-theoretic tools for the analysis of neural encoding. The multielectrode method allowed for the comparison of tuning functions among simultaneously recorded cells. During tracking, MI neurons showed heterogeneity of position and velocity coding, with markedly different temporal dynamics for each. Velocity-tuned neurons were approximately sinusoidally tuned for direction, with linear speed scaling; other cells showed sinusoidal tuning for position, with linear scaling by distance. Velocity encoding led behavior by about 100 ms for most cells, whereas position tuning was more broadly distributed, with leads and lags suggestive of both feedforward and feedback coding. Individual cells encoded velocity and position weakly, with comparable amounts of information about each. Linear regression methods confirmed that random, 2-D hand trajectories can be reconstructed from the firing of small ensembles of randomly selected neurons (3-19 cells) within the MI arm area. These findings demonstrate that MI carries information about evolving hand trajectory during visually guided pursuit tracking, including information about arm position both during and after its specification. However, the reconstruction methods used here capture only the low-frequency components of movement during the PTT. Hand motion signals appear to be represented as a distributed code in which diverse information about position and velocity is available within small regions of MI.

Wider dissemination and testing of computational models are crucial to the field of computational neuroscience. Databases are being developed to meet this need. ModelDB is a web-accessible database for convenient entry, retrieval, and running of published models on different platforms. This article provides a guide to entering a new model into ModelDB.

A number of recent methods developed for automatic classification of multiunit neural activity rely on a gaussian model of the variability of individual waveforms and the statistical methods of gaussian mixture decomposition. Recent evidence has shown that the gaussian model does not accurately capture the multivariate statistics of the waveform samples' distribution. We present further data demonstrating non-gaussian statistics, and show that the multivariate t-distribution, a wide-tailed family of distributions, provides a significantly better fit to the true statistics. We introduce an adaptation of a new Expectation-Maximization (EM) based competitive mixture decomposition algorithm and show that it efficiently and reliably performs mixture decomposition of t-distributions. Our algorithm determines the number of units in multiunit neural recordings, even in the presence of significant noise contamination resulting from random threshold crossings and overlapping spikes.

Some people hear voices that others do not, but only some of those people seek treatment. Using a Pavlovian learning task, we induced conditioned hallucinations in four groups of people who differed orthogonally in their voice-hearing and treatment-seeking statuses. People who hear voices were significantly more susceptible to the effect. Using functional neuroimaging and computational modeling of perception, we identified processes that differentiated voice-hearers from non–voice-hearers and treatment-seekers from non–treatment-seekers and characterized a brain circuit that mediated the conditioned hallucinations. These data demonstrate the profound and sometimes pathological impact of top-down cognitive processes on perception and may represent an objective means to discern people with a need for treatment from those without.

Evidence from imaging and anatomical studies suggests that the midcingulate cortex (MCC) is a dynamic hub lying at the interface of affect and cognition. In particular, this neural system appears to integrate information about conflict and punishment in order to optimize behavior in the face of action-outcome uncertainty. In a series of meta-analyses, we show how recent human electrophysiological research provides compelling evidence that frontal-midline theta signals reflecting MCC activity are moderated by anxiety and predict adaptive behavioral adjustments. These findings underscore the importance of frontal theta activity to a broad spectrum of control operations. We argue that frontal-midline theta provides a neurophysiologically plausible mechanism for optimally adjusting behavior to uncertainty, a hallmark of situations that elicit anxiety and demand cognitive control. These observations compel a new perspective on the mechanisms guiding motivated learning and behavior and provide a framework for understanding the role of the MCC in temperament and psychopathology.

Integration of multiple sensory cues is essential for precise and accurate perception and behavioral performance, yet the reliability of sensory signals can vary across modalities and viewing conditions. Human observers typically employ the optimal strategy of weighting each cue in proportion to its reliability, but the neural basis of this computation remains poorly understood. We trained monkeys to perform a heading discrimination task from visual and vestibular cues, varying cue reliability randomly. The monkeys appropriately placed greater weight on the more reliable cue, and population decoding of neural responses in the dorsal medial superior temporal area closely predicted behavioral cue weighting, including modest deviations from optimality. We found that the mathematical combination of visual and vestibular inputs by single neurons is generally consistent with recent theories of optimal probabilistic computation in neural circuits. These results provide direct evidence for ...

Schizophrenia (SZ) and autism spectrum disorders (ASD) are characterised by marked language deficits, but it is not clear how these arise from gene mutations associated with the disorders. Our goal is to narrow the gap between SZ and ASD and, ultimately, give support to the view that they represent abnormal (but related) ontogenetic itineraries for the human faculty of language. We will focus on the distinctive oscillatory profiles of the SZ and ASD brains, in turn using these insights to refine our understanding of how the brain implements linguistic computations by exploring a novel model of linguistic feature-set composition. We will argue that brain rhythms constitute the best route to interpreting language deficits in both conditions and mapping them to neural dysfunction and risk alleles of the genes. Importantly, candidate genes for SZ and ASD are overrepresented among the gene sets believed to be important for language evolution. This translational effort may help develop an understanding of the aetiology of SZ and ASD and their high prevalence among modern populations.

Schizophrenia is a major mental illness that has a great impact on patients and their environment. One of the difficulties in proposing models for schizophrenia is the complexity and heterogeneity of the illness. There are three main types of symptoms: cognitive, negative and positive. The cognitive symptoms of schizophrenia include distractibility, poor attention, working-memory deficits and the dysexecutive syndrome 1-3 . The negative symptoms include apathy, poor rapport, lack of spontaneity, motor retardation, disturbance of volition, blunted affect, and emotional withdrawal and passive behaviour 1,3 . The positive symptoms of schizophrenia include bizarre trains of thoughts, hallucinations and delusions 1,3 . Theoretical studies using neural-network simulations have started to address how this disorder can have such divergent symptoms. Neural-network models have become increasingly sophisticated. Some current models include neurons with ion channels that open during synaptic activity, which involves particular types of excitatory and inhibitory receptors with biologically realistic time constants. The neurons in such models integrate the effects of the synaptic currents: if a threshold of depolarization is reached, they fire a spike that is transmitted to the synapses on the other neurons in the network. This 'integrate-and-fire' approach (and related approaches) allows the dynamical properties and the stability of the whole neural network to be investigated . For example, researchers can investigate the effects of alterations in ion channels (for instance, in the currents that they pass) on the operation of the whole neural network in, for example, maintaining a short-term memory 7-9 , maintaining attention and initiating action , or in decision making . Given that the cognitive symptoms of schizophrenia include changes in these types of behaviour, these neural-network models are potentially useful in trying to understand the symptoms of schizophrenia. In this Review we show how attractor networks are important for short-term memory and attention, and then how the random firing of neurons can influence the stability of these networks by introducing stochastic noise. We show how some of the different clinical symptoms of schizophrenia could be related to changes in the stability of such networks. Reduced stability of attractor networks can produce a reduced signal-to-noise ratiofor example, when a short-term memory is not always maintained and so the performance is unpredictable and, in this sense, noisy. We relate this to the concept that there is a reduced cortical signal-to-noise ratio in schizophrenia 18 . We then consider how alterations in dopamine in the prefrontal cortex could influence the stability and, thus, the signal-to-noise ratio of attractor networks in the prefrontal cortex and thereby produce some of the cognitive symptoms of schizophrenia.

Action potentials elicited in the axon actively back-propagate into the dendritic tree. During this process their amplitudes can be modulated by internal and external factors. We used a compartmental model of a hippocampal CA1 pyramidal neuron to illustrate how this modulation could depend on (1) the properties of an A-type K+ conductance that is expressed at high density in hippocampal dendrites and (2) the relative timing of synaptic activation. The simulations suggest that the time relationship between pre- and postsynaptic activity could help regulate the amplitude of back-propagating action potentials, especially in the distal portion of the dendritic tree.

We propose a rigorous definition for the term temporal encoding as it is applied to schemes for the representation of information within patterns of neuronal action potentials, and distinguish temporal encoding schemes from those based on window-averaged mean rate encoding. The definition relies on the identification of an encoding time window, defined as the duration of a neuron's spike train assumed to correspond to a single symbol in the neural code. The duration of the encoding time window is dictated by the time scale of the information being encoded. We distinguish between the concepts of the encoding time window and the integration time window, the latter of which is defined as the duration of a stimulus signal that affects the response of the neuron. We note that the duration of the encoding and integration windows might be significantly different. We also present objective, experimentally assessable criteria for identifying neurons and neuronal ensembles that utilize temporal encoding to any significant extent. The definitions and criteria are made rigorous within the contexts of several commonly used analytical approaches, including the stimulus reconstruction analysis technique. Several examples are presented to illustrate the distinctions between and relative capabilities of rate encoding and temporal encoding schemes. We also distinguish our usage of temporal encoding from the term temporal coding, which is commonly used in reference to the representation of information about the timing of events by rate encoding schemes.

We present a biologically plausible model of binocular rivalry consisting of a network of Hodgkin-Huxley type neurons. Our model accounts for the experimentally and psychophysically observed phenomena: (1) it reproduces the distribution of dominance durations seen in both humans and primates, (2) it exhibits a lack of correlation between lengths of successive dominance durations, (3) variation of stimulus strength to one eye influences only the mean dominance duration of the contralateral eye, not the mean dominance duration of the ipsilateral eye, (4) increasing both stimuli strengths in parallel decreases the mean dominance durations. We have also derived a reduced population rate model from our spiking model from which explicit expressions for the dependence of the dominance durations on input strengths are analytically calculated. We also use this reduced model to derive an expression for the distribution of dominance durations seen within an individual.

This paper describes a framework for modeling human activities as temporally structured processes. Our approach is motivated by the inherently hierarchical nature of hu- man activities and the close correspondence between hu- man actions and speech: We model action units using Hid- den Markov Models, much like words in speech. These ac- tion units then form the building blocks to model complex human activities as sentences using an action grammar.
To evaluate our approach, we collected a large dataset of daily cooking activities: The dataset includes a total of 52 participants, each performing a total of 10 cooking ac- tivities in multiple real-life kitchens, resulting in over 77 hours of video footage. We evaluate the HTK toolkit, a state- of-the-art speech recognition engine, in combination with multiple video feature descriptors, for both the recognition of cooking activities (e.g., making pancakes) as well as the semantic parsing of videos into action units (e.g., cracking eggs). Our results demonstrate the benefits of structured temporal generative approaches over existing discrimina- tive approaches in coping with the complexity of human daily life activities.

Cognitive behaviour requires complex context-dependent processing of information that emerges from the links between attentional perceptual processes, working memory and reward-based evaluation of the performed actions. We describe a computational neuroscience theoretical framework which shows how an attentional state held in a short term memory in the prefrontal cortex can by top-down processing influence ventral and dorsal stream cortical areas using biased competition to account for many aspects of visual attention. We also show how within the prefrontal cortex an attentional bias can influence the mapping of sensory inputs to motor outputs, and thus play an important role in decision making. We also show how the absence of expected rewards can switch an attentional bias signal, and thus rapidly and flexibly alter cognitive performance. This theoretical framework incorporates spiking and synaptic dynamics which enable single neuron responses, fMRI activations, psychophysical results, the effects of pharmacological agents, and the effects of damage to parts of the system to be explicitly simulated and predicted. This computational neuroscience framework provides an approach for integrating different levels of investigation of brain function, and for understanding the relations between them. The models also directly address how bottom-up and top-down processes interact in visual cognition, and show how some apparently serial processes reflect the operation of interacting parallel distributed systems. # 1 Covert attention refers to attentional processes that occur in the absence of eye movements. Active visual search using eye movements (overt search) is often necessary in complex natural scenes, because the inferior temporal visual cortex represents primarily what is close to the fovea .

Recent work has examined the estimation of models of stimulus-driven neural activity in which some linear filtering process is followed by a nonlinear, probabilistic spiking stage. We analyze the estimation of one such model for which this nonlinear step is implemented by a noisy, leaky, integrate-and-fire mechanism with a spike-dependent aftercurrent. This model is a biophysically plausible alternative to models with Poisson (memory-less) spiking, and has been shown to effectively reproduce various spiking statistics of neurons in vivo. However, the problem of estimating the model from extracellular spike train data has not been examined in depth. We formulate the problem in terms of maximum likelihood estimation, and show that the computational problem of maximizing the likelihood is tractable. Our main contribution is an algorithm and a proof that this algorithm is guaranteed to find the global optimum with reasonable speed. We demonstrate the effectiveness of our estimator with numerical simulations.

We report a computer simulation of the visuospatial delayed-response experiments of Funahashi et al. (1989), using a firing-rate model that combines intrinsic cellular bistability with the recurrent local network architecture of the neocortex. In our model, the visuospatial working memory is stored in the form of a continuum of network activity profiles that coexist with a spontaneous activity state. These neuronal firing patterns provide a population code for the cue position in a graded manner. We show that neuronal persistent activity and tuning curves of delay-period activity (memory fields) can be generated by an excitatory feedback circuit and recurrent synaptic inhibition. However, if the memory fields are constructed solely by network mechanisms, noise may induce a random drift over time in the encoded cue position, so that the working memory storage becomes unreliable. Furthermore, a "distraction" stimulus presented during the delay period produces a systematic sh...

How does the brain learn those visual features that are relevant for behavior? In this article, we focus on two factors that guide plasticity of visual representations. First, reinforcers cause the global release of diffusive neuromodulatory signals that gate plasticity. Second, attentional feedback signals highlight the chain of neurons between sensory and motor cortex responsible for the selected action. We here propose that the attentional feedback signals guide learning by suppressing plasticity of irrelevant features while permitting the learning of relevant ones. By hypothesizing that sensory signals that are too weak to be perceived can escape from this inhibitory feedback, we bring attentional learning theories and theories that emphasized the importance of neuromodulatory signals into a single, unified framework.

Neurons in the superior colliculus (SC) are known to integrate stimuli of different modalities (e.g., visual and auditory) following specific properties. In this work, we present a mathematical model of the integrative response of SC neurons, in order to suggest a possible physiological mechanism underlying multisensory integration in SC. The model includes three distinct neural areas: two unimodal areas (auditory and visual) are devoted to a topological representation of external stimuli, and communicate via synaptic connections with a third downstream area (in the SC) responsible for multisensory integration. The present simulations show that the model, with a single set of parameters, can mimic various responses to different combinations of external stimuli including the inverse effectiveness, both in terms of multisensory enhancement and contrast, the existence of within- and cross-modality suppression between spatially disparate stimuli, a reduction of network settling time in response to cross-modal stimuli compared with individual stimuli. The model suggests that non-linearities in neural responses and synaptic (excitatory and inhibitory) connections can explain several aspects of multisensory integration.

Plasticity in the brain reaches far beyond a mere changing of synaptic strengths. Recent timelapse imaging in the living brain reveals ongoing structural plasticity by forming or breaking of synapses, motile spines, and re-routing of axonal branches in the developing and adult brain. Some forms of structural plasticity do not follow Hebbian- or anti-Hebbian paradigms of plasticity but rather appear to contribute to the homeostasis of network activity. Four decades of lesion studies have brought up a wealth of data on the mutual interdependence of neuronal activity, neurotransmitter release and neuronal morphogenesis and network formation. Here, we review these former studies on structural plasticity in the context of recent experimental studies. We compare spontaneous and experience-dependent structural plasticity with lesion-induced (reactive) structural plasticity that occurs during development and in the adult brain. Understanding the principles of neural network reorganization on a structural level is relevant for a deeper understanding of long-term memory formation as well as for the treatment of neurological diseases such as stroke.

The spectrotemporal receptive field (STRF) is a functional descriptor of the linear processing of time-varying acoustic spectra by the auditory system. By cross-correlating sustained neuronal activity with the dynamic spectrum of a spectrotemporally rich stimulus ensemble, one obtains an estimate of the STRF. In this article, the relationship between the spectrotemporal structure of any given stimulus and the quality of the STRF estimate is explored and exploited. Invoking the Fourier theorem, arbitrary dynamic spectra are described as sums of basic sinusoidal components--that is, moving ripples. Accurate estimation is found to be especially reliant on the prominence of components whose spectral and temporal characteristics are of relevance to the auditory locus under study and is sensitive to the phase relationships between components with identical temporal signatures. These and other observations have guided the development and use of stimuli with deterministic dynamic spectra co...

According to reinforcement learning theory of decision making, reward expectation is computed by integrating past rewards with a fixed timescale. In contrast, we found that a wide range of time constants is available across cortical neurons recorded from monkeys performing a competitive game task. By recognizing that reward modulates neural activity multiplicatively, we found that one or two time constants of reward memory can be extracted for each neuron in prefrontal, cingulate and parietal cortex. These timescales ranged from hundreds of milliseconds to tens of seconds, according to a power law distribution, which is consistent across areas and reproduced by a 'reservoir' neural network model. These neuronal memory timescales were weakly, but significantly, correlated with those of monkey's decisions. Our findings suggest a flexible memory system in which neural subpopulations with distinct sets of long or short memory timescales may be selectively deployed according to the task demands. We thank J. Mazer and M.W. Jung for comments on an earlier version of the manuscript, and R. Chaudhuri, M. Harre and J. Murray for discussions.

The extreme complexity of the brain naturally requires mathematical modeling approaches on a large variety of scales; the spectrum ranges from single neuron dynamics over the behavior of groups of neurons to neuronal network activity. Thus, the connection between the microscopic scale (single neuron activity) to macroscopic behavior (emergent behavior of the collective dynamics) and vice versa is a key to understand the brain in its complexity. In this work, we attempt a review of a wide range of approaches, ranging from the modeling of single neuron dynamics to machine learning. The models include biophysical as well as data-driven phenomenological models. The discussed models include Hodgkin– Huxley, FitzHugh–Nagumo, coupled oscillators (Kuramoto oscillators, Rössler oscillators, and the Hindmarsh–Rose neuron), Integrate and Fire, networks of neurons, and neural field equations. In addition to the mathematical models, important mathematical methods in multiscale modeling and reconstruction of the causal connectivity are sketched. The methods include linear and nonlinear tools from statistics, data analysis, and time series analysis up to differential equations, dynamical systems, and bifurcation theory, including Granger causal connectivity analysis, phase synchronization connectivity analysis, principal component analysis (PCA), independent component analysis (ICA), and manifold learning algorithms such as ISOMAP, and diffusion maps and equation-free techniques.

Animal learning is associated with changes in the efficacy of connections between neurons. The rules that govern this plasticity can be tested in neural networks. Rules that train neural networks to map stimuli onto outputs are given by supervised learning and reinforcement learning theories. Supervised learning is efficient but biologically implausible. In contrast, reinforcement learning is biologically plausible but comparatively inefficient. It lacks a mechanism that can identify units at early processing levels that play a decisive role in the stimulus-response mapping. Here we show that this so-called credit assignment problem can be solved by a new role for attention in learning. There are two factors in our new learning scheme that determine synaptic plasticity: (1) a reinforcement signal that is homogeneous across the network and depends on the amount of reward obtained after a trial, and (2) an attentional feedback signal from the output layer that limits plasticity to those units at earlier processing levels that are crucial for the stimulus-response mapping. The new scheme is called attention-gated reinforcement learning (AGREL). We show that it is as efficient as supervised learning in classification tasks. AGREL is biologically realistic and integrates the role of feedback connections, attention effects, synaptic plasticity, and reinforcement learning signals into a coherent framework.

Synaptic plasticity is the capacity of a preexisting connection between two neurons to change in strength as a function of neural activity. Because synaptic plasticity is the major candidate mechanism for learning and memory, the elucidation of its constituting mechanisms is of crucial importance in many aspects of normal and pathological brain function. In particular, a prominent aspect that remains debated is how the plasticity mechanisms, that encompass a broad spectrum of temporal and spatial scales, come to play together in a concerted fashion. Here we review and discuss evidence that pinpoints to a possible non-neuronal, glial candidate for such orchestration: the regulation of synaptic plasticity by astrocytes.

The default mode network (DMN) is believed to subserve the baseline mental activity in humans. Its higher energy consumption compared to other brain networks and its intimate coupling with conscious awareness are both pointing to an unknown over-arching function. Many research streams speak in favor of an evolutionarily adaptive role in envisioning experience to anticipate the future. In the present work, we propose a process model that tries to explain how the DMN may implement continuous evaluation and prediction of the environment to guide behavior. The main purpose of DMN activity, we argue, may be described by Markov decision processes that optimize action policies via value estimates through vicarious trial and error. Our formal perspective on DMN function naturally accommodates as special cases previous interpretations based on (a) predictive coding, (b) semantic associations, and (c) a sen-tinel role. Moreover, this process model for the neural optimization of complex behavior in the DMN offers parsimonious explanations for recent experimental findings in animals and humans. K E Y W O R D S artificial intelligence, human intelligence, systems neuroscience

The question of phenomenal experience, of which perception is a part, has yet to be consistently linked to the mechanistic theories of neural function. The authors of this paper explored the Apical Amplification hypothesis, which predicts that Ca 2+ signals in some pyramidal dendrites of L5 might correlate, and perhaps even cause, the perceptual detection of stimuli.

We describe four different mechanisms that lead to oscillations in a network of two reciprocally inhibitory cells. In two cases (intrinsic release and intrinsic escape) the frequency of the network oscillation is insensitive to the threshold voltage of the synaptic potentials. In the other two cases (synaptic release and synaptic escape) the network frequency is strongly determined by the threshold voltage of the synaptic connections. The distinction between the different mechanisms blurs as the function describing synaptic activation becomes less steep and as the model neurons are removed from the relaxation regime. These mechanisms provide insight into the parameters that control network frequency in motor systems that depend on reciprocal inhibition.

In these companion papers, we study how the interrelated dynamics of sodium and potassium affect the excitability of neurons, the occurrence of seizures, and the stability of persistent states of activity. In this first paper, we construct a mathematical model consisting of a single conductance-based neuron together with intra-and extracellular ion concentration dynamics. We formulate a reduction of this model that permits a detailed bifurcation analysis, and show that the reduced model is a reasonable approximation of the full model. We find that competition between intrinsic neuronal currents, sodium-potassium pumps, glia, and diffusion can produce very slow and largeamplitude oscillations in ion concentrations similar to what is seen physiologically in seizures. Using the reduced model, we identify the dynamical mechanisms that give rise to these phenomena. These models reveal several experimentally testable predictions.

We present a model for the self-organized formation of place cells, head-direction cells, and spatial-view cells in the hippocampal formation based on unsupervised learning on quasi-natural visual stimuli. The model comprises a hierarchy of Slow Feature Analysis (SFA) nodes, which were recently shown to reproduce many properties of complex cells in the early visual system [1]. The system extracts a distributed grid-like representation of position and orientation, which is transcoded into a localized place-field, head-direction, or view representation, by sparse coding. The type of cells that develops depends solely on the relevant input statistics, i.e., the movement pattern of the simulated animal. The numerical simulations are complemented by a mathematical analysis that allows us to accurately predict the output of the top SFA layer.

The richness of perceptual experience, as well as its usefulness for guiding behaviour, depends on the synthesis of information across multiple senses. Recent decades have witnessed a surge in our understanding of how the brain combines sensory cues. Much of this research has been guided by one of two distinct approaches: one is driven primarily by neurophysiological observations, and the other is guided by principles of mathematical psychology and psychophysics. Conflicting results and interpretations have contributed to a conceptual gap between psychophysical and physiological accounts of cue integration, but recent studies of visual-vestibular cue integration have narrowed this gap considerably.