Efficient sampling-based Bayesian Active Learning for synaptic characterization.

Bayesian Active Learning (BAL) is an efficient framework for learning the parameters of a model, in which input stimuli are selected to maximize the mutual information between the observations and the unknown parameters. However, the applicability of BAL to experiments is limited as it requires performing high-dimensional integrations and optimizations in real time. Current methods are either too time consuming, or only applicable to specific models. Here, we propose an Efficient Sampling-Based Bayesian Active Learning (ESB-BAL) framework, which is efficient enough to be used in real-time biological experiments. We apply our method to the problem of estimating the parameters of a chemical synapse from the postsynaptic responses to evoked presynaptic action potentials. Using synthetic data and synaptic whole-cell patch-clamp recordings, we show that our method can improve the precision of model-based inferences, thereby paving the way towards more systematic and efficient experimental designs in physiology.

Correction: Bayesian regression explains how human participants handle parameter uncertainty.

[This corrects the article DOI: 10.1371/journal.pcbi.1007886.].

Learning as filtering: Implications for spike-based plasticity.

Most normative models in computational neuroscience describe the task of learning as the optimisation of a cost function with respect to a set of parameters. However, learning as optimisation fails to account for a time-varying environment during the learning process and the resulting point estimate in parameter space does not account for uncertainty. Here, we frame learning as filtering, i.e., a principled method for including time and parameter uncertainty. We derive the filtering-based learning rule for a spiking neuronal network-the Synaptic Filter-and show its computational and biological relevance. For the computational relevance, we show that filtering improves the weight estimation performance compared to a gradient learning rule with optimal learning rate. The dynamics of the mean of the Synaptic Filter is consistent with spike-timing dependent plasticity (STDP) while the dynamics of the variance makes novel predictions regarding spike-timing dependent changes of EPSP variability. Moreover, the Synaptic Filter explains experimentally observed negative correlations between homo- and heterosynaptic plasticity.

Synaptic plasticity as Bayesian inference.

Learning, especially rapid learning, is critical for survival. However, learning is hard; a large number of synaptic weights must be set based on noisy, often ambiguous, sensory information. In such a high-noise regime, keeping track of probability distributions over weights is the optimal strategy. Here we hypothesize that synapses take that strategy; in essence, when they estimate weights, they include error bars. They then use that uncertainty to adjust their learning rates, with more uncertain weights having higher learning rates. We also make a second, independent, hypothesis: synapses communicate their uncertainty by linking it to variability in postsynaptic potential size, with more uncertainty leading to more variability. These two hypotheses cast synaptic plasticity as a problem of Bayesian inference, and thus provide a normative view of learning. They generalize known learning rules, offer an explanation for the large variability in the size of postsynaptic potentials and make falsifiable experimental predictions.

Identifiability of a Binomial Synapse.

Synapses are highly stochastic transmission units. A classical model describing this stochastic transmission is called the binomial model, and its underlying parameters can be estimated from postsynaptic responses to evoked stimuli. The accuracy of parameter estimates obtained via such a model-based approach depends on the identifiability of the model. A model is said to be structurally identifiable if its parameters can be uniquely inferred from the distribution of its outputs. However, this theoretical property does not necessarily imply practical identifiability. For instance, if the number of observations is low or if the recording noise is high, the model's parameters can only be loosely estimated. Structural identifiability, which is an intrinsic property of a model, has been widely characterized; but practical identifiability, which is a property of both the model and the experimental protocol, is usually only qualitatively assessed. Here, we propose a formal definition for the practical identifiability domain of a statistical model. For a given experimental protocol, this domain corresponds to the set of parameters for which the model is correctly identified as the ground truth compared to a simpler alternative model. Considering a model selection problem instead of a parameter inference problem allows to derive a non-arbitrary criterion for practical identifiability. We apply our definition to the study of neurotransmitter release at a chemical synapse. Our contribution to the analysis of synaptic stochasticity is three-fold: firstly, we propose a quantitative criterion for the practical identifiability of a statistical model, and compute the identifiability domains of different variants of the binomial release model (uni or multi-quantal, with or without short-term plasticity); secondly, we extend the Bayesian Information Criterion (BIC), a classically used tool for model selection, to models with correlated data (which is the case for most models of chemical synapses); finally, we show that our approach allows to perform data free model selection, i.e., to verify if a model used to fit data was indeed identifiable even without access to the data, but having only access to the fitted parameters.

Generalized priority-based model for smartphone screen touches.

The distribution of intervals between human actions such as email posts or keyboard strokes demonstrates distinct properties at short versus long timescales. For instance, at long timescales, which are presumably controlled by complex process such as planning and decision making, it has been shown that those interevent intervals follow a scale-invariant (or power-law) distribution. In contrast, at shorter timescales-which are governed by different processes such as sensorimotor skill-they do not follow the same distribution and we know little about how they relate to the scale-invariant pattern. Here, we analyzed 9 million intervals between smartphone screen touches of 84 individuals which span several orders of magnitudes (from milliseconds to hours). To capture these intervals, we extend a priority-based generative model to smartphone touching events. At short timescale, the model is governed by refractory effects, while at longer timescales, the intertouch intervals are governed by the priority difference between smartphone tasks and other tasks. The flexibility of the model allows us to capture interindividual variations at short and long timescales, while its tractability enables efficient model fitting. According to our model, each individual has a specific power-law exponent which is tightly related to the effective refractory time constant suggesting that motor processes which influence the fast actions are related to the higher cognitive processes governing the longer interevent intervals.

Bayesian regression explains how human participants handle parameter uncertainty.

Accumulating evidence indicates that the human brain copes with sensory uncertainty in accordance with Bayes' rule. However, it is unknown how humans make predictions when the generative model of the task at hand is described by uncertain parameters. Here, we tested whether and how humans take parameter uncertainty into account in a regression task. Participants extrapolated a parabola from a limited number of noisy points, shown on a computer screen. The quadratic parameter was drawn from a bimodal prior distribution. We tested whether human observers take full advantage of the given information, including the likelihood of the quadratic parameter value given the observed points and the quadratic parameter's prior distribution. We compared human performance with Bayesian regression, which is the (Bayes) optimal solution to this problem, and three sub-optimal models, which are simpler to compute. Our results show that, under our specific experimental conditions, humans behave in a way that is consistent with Bayesian regression. Moreover, our results support the hypothesis that humans generate responses in a manner consistent with probability matching rather than Bayesian decision theory.

On the choice of metric in gradient-based theories of brain function.

This is a PLOS Computational Biology Education paper. The idea that the brain functions so as to minimize certain costs pervades theoretical neuroscience. Because a cost function by itself does not predict how the brain finds its minima, additional assumptions about the optimization method need to be made to predict the dynamics of physiological quantities. In this context, steepest descent (also called gradient descent) is often suggested as an algorithmic principle of optimization potentially implemented by the brain. In practice, researchers often consider the vector of partial derivatives as the gradient. However, the definition of the gradient and the notion of a steepest direction depend on the choice of a metric. Because the choice of the metric involves a large number of degrees of freedom, the predictive power of models that are based on gradient descent must be called into question, unless there are strong constraints on the choice of the metric. Here, we provide a didactic review of the mathematics of gradient descent, illustrate common pitfalls of using gradient descent as a principle of brain function with examples from the literature, and propose ways forward to constrain the metric.

Model-Based Inference of Synaptic Transmission.

Synaptic computation is believed to underlie many forms of animal behavior. A correct identification of synaptic transmission properties is thus crucial for a better understanding of how the brain processes information, stores memories and learns. Recently, a number of new statistical methods for inferring synaptic transmission parameters have been introduced. Here we review and contrast these developments, with a focus on methods aimed at inferring both synaptic release statistics and synaptic dynamics. Furthermore, based on recent proposals we discuss how such methods can be applied to data across different levels of investigation: from intracellular paired experiments to in vivo network-wide recordings. Overall, these developments open the window to reliably estimating synaptic parameters in behaving animals.

Publisher Correction: Nonlinear Bayesian filtering and learning: a neuronal dynamics for perception.

A correction to this article has been published and is linked from the HTML version of this paper. The error has been fixed in the paper.

Nonlinear Bayesian filtering and learning: a neuronal dynamics for perception.

The robust estimation of dynamical hidden features, such as the position of prey, based on sensory inputs is one of the hallmarks of perception. This dynamical estimation can be rigorously formulated by nonlinear Bayesian filtering theory. Recent experimental and behavioral studies have shown that animals' performance in many tasks is consistent with such a Bayesian statistical interpretation. However, it is presently unclear how a nonlinear Bayesian filter can be efficiently implemented in a network of neurons that satisfies some minimum constraints of biological plausibility. Here, we propose the Neural Particle Filter (NPF), a sampling-based nonlinear Bayesian filter, which does not rely on importance weights. We show that this filter can be interpreted as the neuronal dynamics of a recurrently connected rate-based neural network receiving feed-forward input from sensory neurons. Further, it captures properties of temporal and multi-sensory integration that are crucial for perception, and it allows for online parameter learning with a maximum likelihood approach. The NPF holds the promise to avoid the 'curse of dimensionality', and we demonstrate numerically its capability to outperform weighted particle filters in higher dimensions and when the number of particles is limited.

A Statistical Model for In Vivo Neuronal Dynamics.

Single neuron models have a long tradition in computational neuroscience. Detailed biophysical models such as the Hodgkin-Huxley model as well as simplified neuron models such as the class of integrate-and-fire models relate the input current to the membrane potential of the neuron. Those types of models have been extensively fitted to in vitro data where the input current is controlled. Those models are however of little use when it comes to characterize intracellular in vivo recordings since the input to the neuron is not known. Here we propose a novel single neuron model that characterizes the statistical properties of in vivo recordings. More specifically, we propose a stochastic process where the subthreshold membrane potential follows a Gaussian process and the spike emission intensity depends nonlinearly on the membrane potential as well as the spiking history. We first show that the model has a rich dynamical repertoire since it can capture arbitrary subthreshold autocovariance functions, firing-rate adaptations as well as arbitrary shapes of the action potential. We then show that this model can be efficiently fitted to data without overfitting. We finally show that this model can be used to characterize and therefore precisely compare various intracellular in vivo recordings from different animals and experimental conditions.

Nerve injury-induced neuropathic pain causes disinhibition of the anterior cingulate cortex.

Neuropathic pain caused by peripheral nerve injury is a debilitating neurological condition of high clinical relevance. On the cellular level, the elevated pain sensitivity is induced by plasticity of neuronal function along the pain pathway. Changes in cortical areas involved in pain processing contribute to the development of neuropathic pain. Yet, it remains elusive which plasticity mechanisms occur in cortical circuits. We investigated the properties of neural networks in the anterior cingulate cortex (ACC), a brain region mediating affective responses to noxious stimuli. We performed multiple whole-cell recordings from neurons in layer 5 (L5) of the ACC of adult mice after chronic constriction injury of the sciatic nerve of the left hindpaw and observed a striking loss of connections between excitatory and inhibitory neurons in both directions. In contrast, no significant changes in synaptic efficacy in the remaining connected pairs were found. These changes were reflected on the network level by a decrease in the mEPSC and mIPSC frequency. Additionally, nerve injury resulted in a potentiation of the intrinsic excitability of pyramidal neurons, whereas the cellular properties of interneurons were unchanged. Our set of experimental parameters allowed constructing a neuronal network model of L5 in the ACC, revealing that the modification of inhibitory connectivity had the most profound effect on increased network activity. Thus, our combined experimental and modeling approach suggests that cortical disinhibition is a fundamental pathological modification associated with peripheral nerve damage. These changes at the cortical network level might therefore contribute to the neuropathic pain condition.

Matching recall and storage in sequence learning with spiking neural networks.

Storing and recalling spiking sequences is a general problem the brain needs to solve. It is, however, unclear what type of biologically plausible learning rule is suited to learn a wide class of spatiotemporal activity patterns in a robust way. Here we consider a recurrent network of stochastic spiking neurons composed of both visible and hidden neurons. We derive a generic learning rule that is matched to the neural dynamics by minimizing an upper bound on the Kullback-Leibler divergence from the target distribution to the model distribution. The derived learning rule is consistent with spike-timing dependent plasticity in that a presynaptic spike preceding a postsynaptic spike elicits potentiation while otherwise depression emerges. Furthermore, the learning rule for synapses that target visible neurons can be matched to the recently proposed voltage-triplet rule. The learning rule for synapses that target hidden neurons is modulated by a global factor, which shares properties with astrocytes and gives rise to testable predictions.

A triplet spike-timing-dependent plasticity model generalizes the Bienenstock-Cooper-Munro rule to higher-order spatiotemporal correlations.

Synaptic strength depresses for low and potentiates for high activation of the postsynaptic neuron. This feature is a key property of the Bienenstock-Cooper-Munro (BCM) synaptic learning rule, which has been shown to maximize the selectivity of the postsynaptic neuron, and thereby offers a possible explanation for experience-dependent cortical plasticity such as orientation selectivity. However, the BCM framework is rate-based and a significant amount of recent work has shown that synaptic plasticity also depends on the precise timing of presynaptic and postsynaptic spikes. Here we consider a triplet model of spike-timing-dependent plasticity (STDP) that depends on the interactions of three precisely timed spikes. Triplet STDP has been shown to describe plasticity experiments that the classical STDP rule, based on pairs of spikes, has failed to capture. In the case of rate-based patterns, we show a tight correspondence between the triplet STDP rule and the BCM rule. We analytically demonstrate the selectivity property of the triplet STDP rule for orthogonal inputs and perform numerical simulations for nonorthogonal inputs. Moreover, in contrast to BCM, we show that triplet STDP can also induce selectivity for input patterns consisting of higher-order spatiotemporal correlations, which exist in natural stimuli and have been measured in the brain. We show that this sensitivity to higher-order correlations can be used to develop direction and speed selectivity.

STDP in Adaptive Neurons Gives Close-To-Optimal Information Transmission.

Spike-frequency adaptation is known to enhance the transmission of information in sensory spiking neurons by rescaling the dynamic range for input processing, matching it to the temporal statistics of the sensory stimulus. Achieving maximal information transmission has also been recently postulated as a role for spike-timing-dependent plasticity (STDP). However, the link between optimal plasticity and STDP in cortex remains loose, as does the relationship between STDP and adaptation processes. We investigate how STDP, as described by recent minimal models derived from experimental data, influences the quality of information transmission in an adapting neuron. We show that a phenomenological model based on triplets of spikes yields almost the same information rate as an optimal model specially designed to this end. In contrast, the standard pair-based model of STDP does not improve information transmission as much. This result holds not only for additive STDP with hard weight bounds, known to produce bimodal distributions of synaptic weights, but also for weight-dependent STDP in the context of unimodal but skewed weight distributions. We analyze the similarities between the triplet model and the optimal learning rule, and find that the triplet effect is an important feature of the optimal model when the neuron is adaptive. If STDP is optimized for information transmission, it must take into account the dynamical properties of the postsynaptic cell, which might explain the target-cell specificity of STDP. In particular, it accounts for the differences found in vitro between STDP at excitatory synapses onto principal cells and those onto fast-spiking interneurons.

Synapses with short-term plasticity are optimal estimators of presynaptic membrane potentials.

The trajectory of the somatic membrane potential of a cortical neuron exactly reflects the computations performed on its afferent inputs. However, the spikes of such a neuron are a very low-dimensional and discrete projection of this continually evolving signal. We explored the possibility that the neuron's efferent synapses perform the critical computational step of estimating the membrane potential trajectory from the spikes. We found that short-term changes in synaptic efficacy can be interpreted as implementing an optimal estimator of this trajectory. Short-term depression arose when presynaptic spiking was sufficiently intense as to reduce the uncertainty associated with the estimate; short-term facilitation reflected structural features of the statistics of the presynaptic neuron such as up and down states. Our analysis provides a unifying account of a powerful, but puzzling, form of plasticity.

STDP in Oscillatory Recurrent Networks: Theoretical Conditions for Desynchronization and Applications to Deep Brain Stimulation.

Highly synchronized neural networks can be the source of various pathologies such as Parkinson's disease or essential tremor. Therefore, it is crucial to better understand the dynamics of such networks and the conditions under which a high level of synchronization can be observed. One of the key factors that influences the level of synchronization is the type of learning rule that governs synaptic plasticity. Most of the existing work on synchronization in recurrent networks with synaptic plasticity are based on numerical simulations and there is a clear lack of a theoretical framework for studying the effects of various synaptic plasticity rules. In this paper we derive analytically the conditions for spike-timing dependent plasticity (STDP) to lead a network into a synchronized or a desynchronized state. We also show that under appropriate conditions bistability occurs in recurrent networks governed by STDP. Indeed, a pathological regime with strong connections and therefore strong synchronized activity, as well as a physiological regime with weaker connections and lower levels of synchronization are found to coexist. Furthermore, we show that with appropriate stimulation, the network dynamics can be pushed to the low synchronization stable state. This type of therapeutical stimulation is very different from the existing high-frequency stimulation for deep brain stimulation since once the stimulation is stopped the network stays in the low synchronization regime.

Optimality model of unsupervised spike-timing-dependent plasticity: synaptic memory and weight distribution.

We studied the hypothesis that synaptic dynamics is controlled by three basic principles: (1) synapses adapt their weights so that neurons can effectively transmit information, (2) homeostatic processes stabilize the mean firing rate of the postsynaptic neuron, and (3) weak synapses adapt more slowly than strong ones, while maintenance of strong synapses is costly. Our results show that a synaptic update rule derived from these principles shares features, with spike-timing-dependent plasticity, is sensitive to correlations in the input and is useful for synaptic memory. Moreover, input selectivity (sharply tuned receptive fields) of postsynaptic neurons develops only if stimuli with strong features are presented. Sharply tuned neurons can coexist with unselective ones, and the distribution of synaptic weights can be unimodal or bimodal. The formulation of synaptic dynamics through an optimality criterion provides a simple graphical argument for the stability of synapses, necessary for synaptic memory.

Triplets of spikes in a model of spike timing-dependent plasticity.

Classical experiments on spike timing-dependent plasticity (STDP) use a protocol based on pairs of presynaptic and postsynaptic spikes repeated at a given frequency to induce synaptic potentiation or depression. Therefore, standard STDP models have expressed the weight change as a function of pairs of presynaptic and postsynaptic spike. Unfortunately, those paired-based STDP models cannot account for the dependence on the repetition frequency of the pairs of spike. Moreover, those STDP models cannot reproduce recent triplet and quadruplet experiments. Here, we examine a triplet rule (i.e., a rule which considers sets of three spikes, i.e., two pre and one post or one pre and two post) and compare it to classical pair-based STDP learning rules. With such a triplet rule, it is possible to fit experimental data from visual cortical slices as well as from hippocampal cultures. Moreover, when assuming stochastic spike trains, the triplet learning rule can be mapped to a Bienenstock-Cooper-Munro learning rule.