Recent years have seen a growing interest in inhibitory interneurons and their circuits. A striking property of cortical inhibition is how tightly it balances excitation. Inhibitory currents not only match excitatory currents on average, but track them on a millisecond time scale, whether they are caused by external stimuli or spontaneous fluctuations. We review, together with experimental evidence, recent theoretical approaches that investigate the advantages of such tight balance for coding and computation. These studies suggest a possible revision of the dominant view that neurons represent information with firing rates corrupted by Poisson noise. Instead, tight excitatory/inhibitory balance may be a signature of a highly cooperative code, orders of magnitude more precise than a Poisson rate code. Moreover, tight balance may provide a template that allows cortical neurons to construct high-dimensional population codes and learn complex functions of their inputs.

Two observations about the cortex have puzzled neuroscientists for a long time. First, neural responses are highly variable. Second, the level of excitation and inhibition received by each neuron is tightly balanced at all times. Here, we demonstrate that both properties are necessary consequences of neural networks that represent information efficiently in their spikes. We illustrate this insight with spiking networks that represent dynamical variables. Our approach is based on two assumptions: We assume that information about dynamical variables can be read out linearly from neural spike trains, and we assume that neurons only fire a spike if that improves the representation of the dynamical variables. Based on these assumptions, we derive a network of leaky integrate-and-fire neurons that is able to implement arbitrary linear dynamical systems. We show that the membrane voltage of the neurons is equivalent to a prediction error about a common population-level signal. Among other things, our approach allows us to construct an integrator network of spiking neurons that is robust against many perturbations. Most importantly, neural variability in our networks cannot be equated to noise. Despite exhibiting the same single unit properties as widely used population code models (e.g. tuning curves, Poisson distributed spike trains), balanced networks are orders of magnitudes more reliable. Our approach suggests that spikes do matter when considering how the brain computes, and that the reliability of cortical representations could have been strongly underestimated.

The ventral intraparietal area (VIP) receives converging inputs from visual, somatosensory, auditory and vestibular systems that use diverse reference frames to encode sensory information. A key issue is how VIP combines those inputs together. We mapped the visual and tactile receptive fields of multimodal VIP neurons in macaque monkeys trained to gaze at three different stationary targets. Tactile receptive fields were found to be encoded into a single somatotopic, or head-centered, reference frame, whereas visual receptive fields were widely distributed between eye- to head-centered coordinates. These findings are inconsistent with a remapping of all sensory modalities in a common frame of reference. Instead, they support an alternative model of multisensory integration based on multidirectional sensory predictions (such as predicting the location of a visual stimulus given where it is felt on the skin and vice versa). This approach can also explain related findings in other multimodal areas.

We argue that current theories of multisensory representations are inconsistent with the existence of a large proportion of multimodal neurons with gain fields and partially shifting receptive fields. Moreover, these theories do not fully resolve the recoding and statistical issues involved in multisensory integration. An alternative theory, which we have recently developed and review here, has important implications for the idea of 'frame of reference' in neural spatial representations. This theory is based on a neural architecture that combines basis functions and attractor dynamics. Basis function units are used to solve the recoding problem, whereas attractor dynamics are used for optimal statistical inferences. This architecture accounts for gain fields and partially shifting receptive fields, which emerge naturally as a result of the network connectivity and dynamics.

The brain represents sensory and motor variables through the activity of large populations of neurons. It is not understood how the nervous system computes with these population codes, given that individual neurons are noisy and thus unreliable. We focus here on two general types of computation, function approximation and cue integration, as these are powerful enough to handle a range of tasks, including sensorimotor transformations, feature extraction in sensory systems and multisensory integration. We demonstrate that a particular class of neural networks, basis function networks with multidimensional attractors, can perform both types of computation optimally with noisy neurons. Moreover, neurons in the intermediate layers of our model show response properties similar to those observed in several multimodal cortical areas. Thus, basis function networks with multidimensional attractors may be used by the brain to compute efficiently with population codes.

Many sensory and motor variables are encoded in the nervous system by the activities of large populations of neurons with bell-shaped tuning curves. Extracting information from these population codes is difficult because of the noise inherent in neuronal responses. In most cases of interest, maximum likelihood (ML) is the best read-out method and would be used by an ideal observer. Using simulations and analysis, we show that a close approximation to ML can be implemented in a biologically plausible model of cortical circuitry. Our results apply to a wide range of nonlinear activation functions, suggesting that cortical areas may, in general, function as ideal observers of activity in preceding areas.

The cochlear implant (CI) is a neuroprosthesis that allows profoundly deaf patients to recover speech intelligibility. This recovery goes through long-term adaptative processes to build coherent percepts from the coarse information delivered by the implant. Here we analyzed the longitudinal postimplantation evolution of word recognition in a large sample of CI users in unisensory (visual or auditory) and bisensory (visuoauditory) conditions. We found that, despite considerable recovery of auditory performance during the first year postimplantation, CI patients maintain a much higher level of word recognition in speechreading conditions compared with normally hearing subjects, even several years after implantation. Consequently, we show that CI users present higher visuoauditory performance when compared with normally hearing subjects with similar auditory stimuli. This better performance is not only due to greater speechreading performance, but, most importantly, also due to a greater capacity to integrate visual input with the distorted speech signal. Our results suggest that these behavioral changes in CI users might be mediated by a reorganization of the cortical network involved in speech recognition that favors a more specific involvement of visual areas. Furthermore, they provide crucial indications to guide the rehabilitation of CI patients by using visually oriented therapeutic strategies.cochlear implant ͉ deafness ͉ multisensory integration ͉ speech comprehension D espite the apparent division between sensory modalities from the receptors to high cortical levels, we can simultaneously integrate visual and auditory signals resulting in qualitative percepts distinct from those derived from a single unisensory stimulus (1, 2). Furthermore, in cases of precise temporal or spatial congruency between the bisensory stimuli, multisensory integration is expressed at the behavioral level by perceptual improvements by reducing ambiguity (3, 4) and at the neuronal level by enhancing neuronal activity (5). Multisensory integration is also essential for speech recognition, which is based on the simultaneous integration of visual information derived from lip movements and auditory cues produced by the talker (6). The McGurk effect, in which a mismatch between the visual and auditory speech signals is artificially introduced, reveals that the visual information derived from lip movements can strongly influence our auditory perception (7). Although we might not be aware of the relevance of the visual cues for normal speech recognition, the influence of vision becomes convincingly apparent when the auditory information is embedded in noise. In degraded auditory conditions, the visuoauditory presentation leads to higher performance of recognition, when compared with the auditory alone stimulation (8, 9), in a mechanism that mimics an improvement in the acoustic signal-to-noise ratio (SNR) (10).In normally hearing (NH) subjects, although speechreading performance is very low, the association during development between the...

We show that the dynamics of spiking neurons can be interpreted as a form of Bayesian inference in time. Neurons that optimally integrate evidence about events in the external world exhibit properties similar to leaky integrate-and-fire neurons with spike-dependent adaptation and maximally respond to fluctuations of their input. Spikes signal the occurrence of new information-what cannot be predicted from the past activity. As a result, firing statistics are close to Poisson, albeit providing a deterministic representation of probabilities.