Summary Spontaneous waves of activity propagating across large cortical areas may play important roles in sensory processing and circuit refinement. However, whether these waves are in turn shaped by sensory experience remains unclear. Here we report that visually evoked cortical activity reverberates in subsequent spontaneous waves. Voltage-sensitive dye imaging in rat visual cortex showed that following repetitive presentation of a given visual stimulus, spatiotemporal activity patterns resembling the evoked response appear more frequently in the spontaneous waves. This effect is specific to the response pattern evoked by the repeated stimulus, and it persists for several minutes without further visual stimulation. Such wave-mediated reverberation could contribute to short-term memory and help to consolidate the transient effects of recent sensory experience into long-lasting cortical modifications.

Deep learning models stand for a new learning paradigm in artificial intelligence (AI) and machine learning. Recent breakthrough results in image analysis and speech recognition have generated a massive interest in this field because also applications in many other domains providing big data seem possible. On a downside, the mathematical and computational methodology underlying deep learning models is very challenging, especially for interdisciplinary scientists. For this reason, we present in this paper an introductory review of deep learning approaches including Deep Feedforward Neural Networks (D-FFNN), Convolutional Neural Networks (CNNs), Deep Belief Networks (DBNs), Autoencoders (AEs), and Long Short-Term Memory (LSTM) networks. These models form the major core architectures of deep learning models currently used and should belong in any data scientist's toolbox. Importantly, those core architectural building blocks can be composed flexibly-in an almost Lego-like manner-to build new application-specific network architectures. Hence, a basic understanding of these network architectures is important to be prepared for future developments in AI.

A central hypothesis concerning sensory processing is that the neuronal circuits are specifically adapted to represent natural stimuli efficiently. Here we show a novel effect in cortical coding of natural images. Using spike-triggered average or spike-triggered covariance analyses, we first identified the visual features selectively represented by each cortical neuron from its responses to natural images. We then measured the neuronal sensitivity to these features when they were present in either natural images or random stimuli. We found that in the responses of complex cells, but not of simple cells, the sensitivity was markedly higher for natural images than for random stimuli. Such elevated sensitivity leads to increased detectability of the visual features and thus an improved cortical representation of natural scenes. Interestingly, this effect is due not to the spatial power spectra of natural images, but to their phase regularities. These results point to a distinct visual-coding strategy that is mediated by contextual modulation of cortical responses tuned to the spatial-phase structure of natural scenes.

No abstract

An essential step in understanding visual processing is to characterize the neuronal receptive fields (RFs) at each stage of the visual pathway. However, RF characterization beyond simple cells in the primary visual cortex (V1) remains a major challenge. Recent application of spike-triggered covariance (STC) analysis has greatly facilitated characterization of complex cell RFs in anesthetized animals. Here we apply STC to RF characterization in awake monkey V1. We found up to nine subunits for each cell, including one or two dominant excitatory subunits as described by the standard model, along with additional excitatory and suppressive subunits with weaker contributions. Compared with the dominant subunits, the nondominant excitatory subunits prefer similar orientations and spatial frequencies but have larger spatial envelopes. They contribute to response invariance to small changes in stimulus orientation, position, and spatial frequency. In contrast, the suppressive subunits are tuned to orientations 45°-90°differ-ent from the excitatory subunits, which may underlie crossorientation suppression. Together, the excitatory and suppressive subunits form a compact description of RFs in awake monkey V1, allowing prediction of the responses to arbitrary visual stimuli.T he response properties of primary visual cortical (V1) neurons have been studied extensively over the past several decades. In the standard model, a simple cell receptive field (RF) consists of alternating ON and OFF subregions, which directly correspond to the orientation and spatial-frequency tuning of the cell (1, 2). Complex cells exhibit orientation and spatial-frequency tuning similar to simple cells, but they are insensitive to the contrast polarity and stimulus position within the RF. The energy model for complex cell RF consists of a pair of simple-cell-like subunits with the same orientation and spatial-frequency tuning but different ON/OFF phases (3, 4). This model accounts for the phase invariance as well as stimulus selectivity of complex cells.To validate such RF models and to predict the neuronal responses to arbitrary visual stimuli, it is necessary to measure the RF structure quantitatively. For simple cells, spike-triggered average (STA) has been used effectively to estimate their RFs from the responses to sparse noise (5) or white noise (6). For complex cells, however, because the outputs of different RF subunits are combined nonlinearly, these subunits cannot be estimated by STA. In previous studies, complex cell RFs have been studied by measuring the nonlinear interaction between paired stimuli (3,7,8). Another method used in recent studies is spike-triggered covariance (STC) analysis (9, 10). Instead of averaging all of the stimuli preceding spikes (as in STA), in STC analysis one computes the covariance matrix of the spike-triggered stimulus ensemble and identifies the eigenvectors with eigenvalues significantly different from those of the entire stimulus ensemble. This method can reveal stimulus features that drive the neuron ...