Skilled behavior often displays signatures of Bayesian inference. In order for the brain to implement the required computations, neuronal activity must carry accurate information about the uncertainty of sensory inputs. Two major approaches have been proposed to study neuronal representations of uncertainty. The first one, the Bayesian decoding approach, aims primarily at decoding the posterior probability distribution of the stimulus from population activity using Bayes' rule, and indirectly yields uncertainty estimates as a by-product. The second one, which we call the correlational approach, searches for specific features of neuronal activity (such as tuning-curve width and maximum firing-rate) which correlate with uncertainty. To compare these two approaches, we derived a new normative model of sound source localization by Interaural Time Difference (ITD), that reproduces a wealth of behavioral and neural observations. We found that several features of neuronal activity correlated with uncertainty on average, but none provided an accurate estimate of uncertainty on a trial-by-trial basis, indicating that the correlational approach may not reliably identify which aspects of neuronal responses represent uncertainty. In contrast, the Bayesian decoding approach reveals that the activity pattern of the entire population was required to reconstruct the trial-to-trial posterior distribution with Bayes' rule. These results suggest that uncertainty is unlikely to be represented in a single feature of neuronal activity, and highlight the importance of using a Bayesian decoding approach when exploring the neural basis of uncertainty.

Biological brains are inherently limited in their capacity to process and store information, but are nevertheless capable of solving complex tasks with apparent ease. Intelligent behavior is related to these limitations, since resource constraints drive the need to generalize and assign importance differentially to features in the environment or memories of past experiences. Recently, there have been parallel efforts in reinforcement learning and neuroscience to understand strategies adopted by artificial and biological agents to circumvent limitations in information storage. However, the two threads have been largely separate. In this article, we propose a dynamical framework to maximize expected reward under constraints of limited resources, which we implement with a cost function that penalizes precise representations of action-values in memory, each of which may vary in its precision. We derive from first principles an algorithm, Dynamic Resource Allocator (DRA), which we apply to two standard tasks in reinforcement learning and a model-based planning task, and find that it allocates more resources to items in memory that have a higher impact on cumulative rewards. Moreover, DRA learns faster when starting with a higher resource budget than what it eventually allocates for performing well on tasks, which may explain why frontal cortical areas in biological brains appear more engaged in early stages of learning before settling to lower asymptotic levels of activity. Our work provides a normative solution to the problem of learning how to allocate costly resources to a collection of uncertain memories in a manner that is capable of adapting to changes in the environment.

Recent theoretical and experimental results suggest that the dopamine system implements distributional temporal difference backups, allowing learning of the entire distributions of the long-run values of states rather than just their expected values. However, the distributional codes explored so far rely on a complex imputation step which crucially relies on spatial non-locality: in order to compute reward prediction errors, units must know not only their own state but also the states of the other units. It is far from clear how these steps could be implemented in realistic neural circuits. Here, we introduce the Laplace code: a local temporal difference code for distributional reinforcement learning that is representationally powerful and computationally straightforward. The code decomposes value distributions and prediction errors across three separated dimensions: reward magnitude (related to distributional quantiles), temporal discounting (related to the Laplace transform of future rewards) and time horizon (related to eligibility traces). Besides lending itself to a local learning rule, the decomposition recovers the temporal evolution of the immediate reward distribution, indicating all possible rewards at all future times. This increases representational capacity and allows for temporally-flexible computations that immediately adjust to changing horizons or discount factors.

Studies of neuron-behaviour correlation and causal manipulation have long been used separately to understand the neural basis of perception. Yet these approaches sometimes lead to drastically conflicting conclusions about the functional role of brain areas. Theories that focus only on choice-related neuronal activity cannot reconcile those findings without additional experiments involving large-scale recordings to measure interneuronal correlations. By expanding current theories of neural coding and incorporating results from inactivation experiments, we demonstrate here that it is possible to infer decoding weights of different brain areas at a coarse scale without precise knowledge of the correlation structure. We apply this technique to neural data collected from two different cortical areas in macaque monkeys trained to perform a heading discrimination task. We identify two opposing decoding schemes, each consistent with data depending on the nature of correlated noise. Our theory makes specific testable predictions to distinguish these scenarios experimentally without requiring measurement of the underlying noise correlations.

Sensory neurons give highly variable responses to stimulation, which can limit the amount of stimulus information available to downstream circuits. Much work has investigated the factors that affect the amount of information encoded in these population responses, leading to insights about the role of covariability among neurons, tuning curve shape, etc. However, the informativeness of neural responses is not the only relevant feature of population codes; of potentially equal importance is how robustly that information propagates to downstream structures. For instance, to quantify the retina's performance, one must consider not only the informativeness of the optic nerve responses, but also the amount of information that survives the spike-generating nonlinearity and noise corruption in the next stage of processing, the lateral geniculate nucleus. Our study identifies the set of covariance structures for the upstream cells that optimize the ability of information to propagate through noisy, nonlinear circuits. Within this optimal family are covariances with "differential correlations", which are known to reduce the information encoded in neural population activities. Thus, covariance structures that maximize information in neural population codes, and those that maximize the ability of this information to propagate, can be very different. Moreover, redundancy is neither necessary nor sufficient to make population codes robust against corruption by noise: redundant codes can be very fragile, and synergistic codes can-in some cases-optimize robustness against noise.

Neural responses are known to be variable. In order to understand how this neural variability constrains behavioral performance, we need to be able to measure the reliability with which a sensory stimulus is encoded in a given population. However, such measures are challenging for two reasons: First, they must take into account noise correlations which can have a large influence on reliability. Second, they need to be as efficient as possible, since the number of trials available in a set of neural recording is usually limited by experimental constraints. Traditionally, cross-validated decoding has been used as a reliability measure, but it only provides a lower bound on reliability and underestimates reliability substantially in small datasets. We show that, if the number of trials per condition is larger than the number of neurons, there is an alternative, direct estimate of reliability which consistently leads to smaller errors and is much faster to compute. The superior performance of the direct estimator is evident both for simulated data and for neuronal population recordings from macaque primary visual cortex. Furthermore we propose generalizations of the direct estimator which measure changes in stimulus encoding across conditions and the impact of correlations on encoding and decoding, typically denoted by Ishuffle and Idiag respectively.

The olfactory system faces a difficult inference problem: it has to determine what odors are present based on the distributed activation of its receptor neurons. Here we derive neural implementations of two approximate inference algorithms that could be used by the brain. One is a variational algorithm (which builds on the work of Beck. et al., 2012), the other is based on sampling. Importantly, we use a more realistic prior distribution over odors than has been used in the past: we use a spike and slab'' prior, for which most odors have zero concentration. After mapping the two algorithms onto neural dynamics, we find that both can infer correct odors in less than 100 ms, although it takes ~500 ms to eliminate false positives. Thus, at the behavioral level, the two algorithms make very similar predictions. However, they make different assumptions about connectivity and neural computations, and make different predictions about neural activity. Thus, they should be distinguishable experimentally. If so, that would provide insight into the mechanisms employed by the olfactory system, and, because the two algorithms use very different coding strategies, that would also provide insight into how networks represent probabilities."

Recent experiments have demonstrated that humans and animals typically reason probabilistically about their environment. This ability requires a neural code that represents probability distributions and neural circuits that are capable of implementing the operations of probabilistic inference. The proposed probabilistic population coding (PPC) framework provides a statistically efficient neural representation of probability distributions that is both broadly consistent with physiological measurements and capable of implementing some of the basic operations of probabilistic inference in a biologically plausible way. However, these experiments and the corresponding neural models have largely focused on simple (tractable) probabilistic computations such as cue combination, coordinate transformations, and decision making. As a result it remains unclear how to generalize this framework to more complex probabilistic computations. Here we address this short coming by showing that a very general approximate inference algorithm known as Variational Bayesian Expectation Maximization can be implemented within the linear PPC framework. We apply this approach to a generic problem faced by any given layer of cortex, namely the identification of latent causes of complex mixtures of spikes. We identify a formal equivalent between this spike pattern demixing problem and topic models used for document classification, in particular Latent Dirichlet Allocation (LDA). We then construct a neural network implementation of variational inference and learning for LDA that utilizes a linear PPC. This network relies critically on two non-linear operations: divisive normalization and super-linear facilitation, both of which are ubiquitously observed in neural circuits. We also demonstrate how online learning can be achieved using a variation of Hebb’s rule and describe an extesion of this work which allows us to deal with time varying and correlated latent causes.

A simple expression for a lower bound of Fisher information is derived for a network of recurrently connected spiking neurons that have been driven to a noise-perturbed steady state. We call this lower bound linear Fisher information, as it corresponds to the Fisher information that can be recovered by a locally optimal linear estimator. Unlike recent similar calculations, the approach used here includes the effects of nonlinear gain functions and correlated input noise and yields a surprisingly simple and intuitive expression that offers substantial insight into the sources of information degradation across successive layers of a neural network. Here, this expression is used to (1) compute the optimal (i.e., information-maximizing) firing rate of a neuron, (2) demonstrate why sharpening tuning curves by either thresholding or the action of recurrent connectivity is generally a bad idea, (3) show how a single cortical expansion is sufficient to instantiate a redundant population code that can propagate across multiple cortical layers with minimal information loss, and (4) show that optimal recurrent connectivity strongly depends on the covariance structure of the inputs to the network.

Several recent models have proposed the use of precise timing of spikes for cortical computation. Such models rely on growing experimental evidence that neurons in the thalamus as well as many primary sensory cortical areas respond to stimuli with remarkable temporal precision. Models of computation based on spike timing, where the output of the network is a function not only of the input but also of an independently initializable internal state of the network, must, however, satisfy a critical constraint: the dynamics of the network should not be sensitive to initial conditions. We have previously developed an abstract dynamical system for networks of spiking neurons that has allowed us to identify the criterion for the stationary dynamics of a network to be sensitive to initial conditions. Guided by this criterion, we analyzed the dynamics of several recurrent cortical architectures, including one from the orientation selectivity literature. Based on the results, we conclude that under conditions of sustained, Poisson-like, weakly correlated, low to moderate levels of internal activity as found in the cortex, it is unlikely that recurrent cortical networks can robustly generate precise spike trajectories, that is, spatiotemporal patterns of spikes precise to the millisecond timescale.

The codes obtained from the responses of large populations of neurons are known as population codes. Several studies have shown that the amount of information conveyed by such codes, and the format of this information, is highly dependent on the pattern of correlations. However, very little is known about the impact of response correlations (as found in actual cortical circuits) on neural coding. To address this problem, we investigated the properties of population codes obtained from motion energy filters, which provide one of the best models for motion selectivity in early visual areas. It is therefore likely that the correlations that arise among energy filters also arise among motion-selective neurons. We adopted an ideal observer approach to analyze filter responses to three sets of images: noisy sine gratings, random dots kinematograms, and images of natural scenes. We report that in our model, the structure of the population code varies with the type of image. We also show that for all sets of images, correlations convey a large fraction of the information: 40% to 90% of the total information. Moreover, ignoring those correlations when decoding leads to considerable information loss-from 50% to 93%, depending on the image type. Finally we show that it is important to consider a large population of motion energy filters in order to see the impact of correlations. Study of pairs of neurons, as is often done experimentally, can underestimate the effect of correlations.

From first principles, we derive a quadratic nonlinear, first-order dynamical system capable of performing exact Bayes-Markov inferences for a wide class of biologically plausible stimulus-dependent patterns of activity while simultaneously providing an online estimate of model performance. This is accomplished by constructing a dynamical system that has solutions proportional to the probability distribution over the stimulus space, but with a constant of proportionality adjusted to provide a local estimate of the probability of the recent observations of stimulus-dependent activity-given model parameters. Next, we transform this exact equation to generate nonlinear equations for the exact evolution of log likelihood and log-likelihood ratios and show that when the input has low amplitude, linear rate models for both the likelihood and the log-likelihood functions follow naturally from these equations. We use these four explicit representations of the probability distribution to argue that, in contrast to the arguments of previous work, the dynamical system for the exact evolution of the likelihood (as opposed to the log likelihood or log-likelihood ratios) not only can be mapped onto a biologically plausible network but is also more consistent with physiological observations.

If the cortex uses spike timing to compute, the timing of the spikes must be robust to perturbations. Based on a recent framework that provides a simple criterion to determine whether a spike sequence produced by a generic network is sensitive to initial conditions, and numerical simulations of a variety of network architectures, we argue within the limits set by our model of the neuron, that it is unlikely that precise sequences of spike timings are used for computation under conditions typically found in the cortex.

Most models of spatial representations in the cortex assume cells with limited receptive fields that are defined in a particular egocen(cid:173) tric frame of reference. However, cells outside of primary sensory cortex are either gain modulated by postural input or partially shifting. We show that solving classical spatial tasks, like sen(cid:173) sory prediction, multi-sensory integration, sensory-motor transfor(cid:173) mation and motor control requires more complicated intermediate representations that are not invariant in one frame of reference. We present an iterative basis function map that performs these spatial tasks optimally with gain modulated and partially shifting units, and tests it against neurophysiological and neuropsycholog(cid:173) ical data. In order to perform an action directed toward an object, it is necessary to have a representation of its spatial location. The brain must be able to use spatial cues coming from different modalities (e.g. vision, audition, touch, proprioception), combine them to infer the position of the object, and compute the appropriate movement. These cues are in different frames of reference corresponding to different sensory or motor modalities. Visual inputs are primarily encoded in retinotopic maps, auditory inputs are encoded in head centered maps and tactile cues are encoded in skin-centered maps. Going from one frame of reference to the other might seem easy. For example, the head-centered position of an object can be approximated by the sum of its retinotopic position and the eye position. However, positions are represented by population codes in the brain, and computing a head-centered map from a retinotopic map is a more complex computation than the underlying sum. Moreover, as we get closer to sensory-motor areas it seems reasonable to assume

Neurophysiologists are often faced with the problem of evaluating the quality of a code for a sensory or motor variable, either to relate it to the performance of the animal in a simple discrimination task or to compare the codes at various stages along the neuronal pathway. One common belief that has emerged from such studies is that sharpening of tuning curves improves the quality of the code, although only to a certain point; sharpening beyond that is believed to be harmful. We show that this belief relies on either problematic technical analysis or improper assumptions about the noise. We conclude that one cannot tell, in the general case, whether narrow tuning curves are better than wide ones; the answer depends critically on the covariance of the noise. The same conclusion applies to other manipulations of the tuning curve profiles such as gain increase.

Coarse codes are widely used throughout the brain to encode sensory and motor variables. Methods designed to interpret these codes, such as population vector analysis, are either inefficient (the variance of the estimate is much larger than the smallest possible variance) or biologically implausible, like maximum likelihood. Moreover, these methods attempt to compute a scalar or vector estimate of the encoded variable. Neurons are faced with a similar estimation problem. They must read out the responses of the presynaptic neurons, but, by contrast, they typically encode the variable with a further population code rather than as a scalar. We show how a nonlinear recurrent network can be used to perform estimation in a near-optimal way while keeping the estimate in a coarse code format. This work suggests that lateral connections in the cortex may be involved in cleaning up uncorrelated noise among neurons representing similar variables.

We present a general encoding-decoding framework for interpreting the activity of a population of units. A standard population code interpretation method, the Poisson model, starts from a description as to how a single value of an underlying quantity can generate the activities of each unit in the population. In casting it in the encoding-decoding framework, we find that this model is too restrictive to describe fully the activities of units in population codes in higher processing areas, such as the medial temporal area. Under a more powerful model, the population activity can convey information not only about a single value of some quantity but also about its whole distribution, including its variance, and perhaps even the certainty the system has in the actual presence in the world of the entity generating this quantity. We propose a novel method for forming such probabilistic interpretations of population codes and compare it to the existing method.

The auditory system of the barn owl contains several spatial maps. In young barn owls raised with optical prisms over their eyes, these auditory maps are shifted to stay in register with the visual map, suggesting that the visual input imposes a frame of reference on the auditory maps. However, the optic tectum, the first site of convergence of visual with auditory information, is not the site of plasticity for the shift of the auditory maps; the plasticity occurs instead in the inferior colliculus, which contains an auditory map and projects into the optic tectum. We explored a model of the owl remapping in which a global reinforcement signal whose delivery is controlled by visual foveation. A hebb learning rule gated by rein(cid:173) forcement learned to appropriately adjust auditory maps. In addi(cid:173) tion, reinforcement learning preferentially adjusted the weights in the inferior colliculus, as in the owl brain, even though the weights were allowed to change throughout the auditory system. This ob(cid:173) servation raises the possibility that the site of learning does not have to be genetically specified, but could be determined by how the learning procedure interacts with the network architecture. 126 Alexandre Pouget, Cedric Deffayet, Te"ence J. Sejnowski c:::======:::::» •

The parietal cortex is thought to represent the egocentric posi(cid:173) tions of objects in particular coordinate systems. We propose an alternative approach to spatial perception of objects in the pari(cid:173) etal cortex from the perspective of sensorimotor transformations. The responses of single parietal neurons can be modeled as a gaus(cid:173) sian function of retinal position multiplied by a sigmoid function of eye position, which form a set of basis functions. We show here how these basis functions can be used to generate receptive fields in either retinotopic or head-centered coordinates by simple linear transformations. This raises the possibility that the parietal cortex does not attempt to compute the positions of objects in a partic(cid:173) ular frame of reference but instead computes a general purpose representation of the retinal location and eye position from which any transformation can be synthesized by direct projection. This representation predicts that hemineglect, a neurological syndrome produced by parietal lesions, should not be confined to egocentric coordinates, but should be observed in multiple frames of reference in single patients, a prediction supported by several experiments. 158 Alexandre Pouget, Terrence J. Sejnowski