Section: New Results

Visual Neuroscience

On the effects on cortical spontaneous activity of the symmetries of the network of pinwheels in visual area V1

Participants : Romain Veltz, Pascal Chossat, Olivier Faugeras.

We consider the problem of describing mathematically the spontaneous activity of V1 by combining several important experimental observations including 1) the organization of the visual cortex into a spatially periodic network of hypercolumns structured around pinwheels, 2) the difference between short-range and long-range intra-cortical connections, the first ones being rather isotropic and producing naturally doubly-periodic patterns by Turing mechanisms, the second one being patchy and 3) the fact that the Turing patterns spontaneously produced by the short-range connections and the network of pinwheels have similar periods. By analyzing the Preferred Orientation (PO) map, we are able to classify all possible singular points of the PO maps (the pinwheels) as having symmetries described by a small subset of the wallpaper groups. We then propose a description of the spontaneous activity of V1 using a classical voltage-based neural field model that features isotropic short-range connectivities modulated by non-isotropic long-range connectivities. A key observation is that, with only short-range connections and because the problem has full translational invariance in this case, a spontaneous doubly-periodic pattern generates a 2-torus in a suitable functional space which persists as a flow-invariant manifold under small perturbations, hence when turning on the long-range connections. Through a complete analysis of the symmetries of the resulting neural field equation and motivated by a numerical investigation of the bifurcations of their solutions, we conclude that the branches of solutions which are stable over an extended set of parameters are those corresponding to patterns with an hexagonal (or nearly hexagonal) symmetry. The question of which patterns persist when turning on the long-range connections is answered by 1) analyzing the remaining symmetries on the perturbed torus and 2) combining this information with the Poincaré-Hopf theorem. We have developed a numerical implementation of the theory that has allowed us to produce the patterns of activities predicted by the theory, the planforms. In particular we generalize the contoured and non-contoured planforms predicted by previous authors and predict the existence of mixed contoured/non-contoured planforms. We also found that these planforms are most likely to be time evolving. This work is available as a preprint [36] and has been submitted to a Journal.

Decoding the retina with the first wave of spikes

Participants : Geoffrey Portelli, John Barrett [Institute of Neuroscience, Medical School, Newcastle University, Newcastle UK] , Evelyne Sernagor [Institute of Neuroscience, Medical School, Newcastle University, Newcastle UK] , Timothée Masquelier [Institut de la Vision, UPMC Université Paris 06, Paris, France] , Pierre Kornprobst [correspondent] .

How a population of retinal ganglion cells (RGCs) encode the visual scene remains an open question. Several coding strategies have been investigated out of which two main views have emerged: considering RGCs as independent encoders or as synergistic encoders, i.e., when the concerted spiking in a RGC population carries more information than the sum of the information contained in the spiking of individual RGCs. Although the RGCs assumed as independent encode the main information, there is currently a growing body of evidence that considering RGCs as synergistic encoders provides complementary and more precise information. Based on salamander retina recordings, it has been suggested  [66] that a code based on differential spike latencies between RGC pairs could be a powerful mechanism. Here, we have tested this hypothesis in the mammalian retina. We recorded responses to stationary gratings from 469 RGCs in 5 mouse retinas. Interestingly, we did not find any RGC pairs exhibiting clear latency correlations (presumably due to the presence of spontaneous activity), showing that individual RGC pairs do not provide sufficient information in our conditions. However considering the whole RGC population, we show that the shape of the wave of first spikes (WFS) successfully encodes for spatial cues. To quantify its coding capabilities, we performed a discrimination task and we showed that the WFS was more robust to the spontaneous firing than the absolute latencies are. We also investigated the impact of a post-processing neural layer. The recorded spikes were fed into an artificial lateral geniculate nucleus (LGN) layer. We found that the WFS is not only preserved but even refined through the LGN-like layer, while classical independent coding strategies become impaired. These findings suggest that even at the level of the retina, the WFS provides a reliable strategy to encode spatial cues.

This work is ongoing and was presented as a poster at CNS 2014. See [26] for more details.

Microsaccades enable efficient synchrony-based visual feature detection by retinal ganglion cells

Participants : Timothée Masquelier [Institut de la Vision, UPMC Université Paris 06, Paris, France] , Geoffrey Portelli, Pierre Kornprobst [correspondent] .

Fixational eye movements are common across vertebrates, yet their functional roles, if any, are debated [70] . To investigate this issue, we exposed the Virtual Retina simulator [84] to natural images, generated realistic drifts and microsaccades [59] , and analyzed the output spike trains of the parvocellular retinal ganglion cells (RGC). We first computed cross-correlograms between pairs of RGC that are strongly excited by the image corresponding to the mean eye position. Not surprisingly, in the absence of eye movements, that is when analyzing the tonic (sustained) response to a static image, these cross-correlograms are flat. Adding some slow drift ($\approx$20 min/s, self- avoiding random walk) creates long timescale (>1s) correlations because both cells tend to have high firing rates for central positions. Adding microsaccades ($\approx$0.5${}^{\circ }$ in 25ms, that is $\approx$20${}^{\circ }$/s) creates short timescale (tens of ms) correlations: cells that are strongly excited at a particular landing location tend to spike synchronously shortly after the landing. What do the patterns of synchronous spikes represent? To investigate this issue, we fed the RGC spike trains to neurons equipped with spike timing-dependent plasticity (STDP) and lateral inhibitory connections [73] . Neurons self-organized, and each one selected a set of afferents that consistently fired synchronously. We then reconstructed the corresponding visual stimuli by convolving the synaptic weight matrices with the RGC receptive fields. In most cases, we could easily recognize what was learned (e.g. a face), and the neuron was selective (e.g. only responded for microsaccades that landed on a face). Without eye movements, or with only the drift, the STDP- based learning failed, because it needs correlations at a timescale roughly matching the STDP time constants [65] . Microsaccades are thus necessary to generate a synchrony-based coding scheme. More specifically, after each microsaccade landing, cells that are strongly excited by the image corresponding to the landing location tend to fire their first spikes synchronously. Patterns of synchronous spikes can be decoded rapidly – as soon as the first spikes are received – by downstream “coincidence detector” neurons, which do not need to know the landing times. Finally, the required connectivity to do so can spontaneously emerge with STDP. As a whole, these results suggest a new role for microsaccades – to enable efficient visual feature learning and detection thanks to synchronization – that differs from other proposals such as time-to-first spike coding with respect to microsaccade landing times.

This work is ongoing and was presented as a poster at CNS 2014.

A new retina-inspired descriptor for image classification

Participants : Cristina Hilario [Pattern Analysis and Computer Vision, PAVIS, Istituto Italiano di Tecnologia, Genova, Italy] , Diego Sona [Pattern Analysis and Computer Vision, PAVIS, Istituto Italiano di Tecnologia, Genova, Italy] , Kartheek Medathati, Pierre Kornprobst [correspondent] .

Recent studies on the visual system reveal that retina is smarter than scientists believed. One low level processing occurring at the retina is feature extraction, becoming an inspiration to build novel image descriptors for image categorization. However only few methods have taken advantage of this idea, such as FREAK descriptor [51] , which consists of a circular grid of a concentric distribution of overlapping receptive fields (RFs) in which average image intensities are compared pairwise. In this work we extended such a descriptor but sticking much more to biological data and models of retina. Each RF in our model is described with a linear-nonlinear model (LN) taking into account inhibitory surrounds with parameters based on biological findings. Based on the activity of retinal ganglion cells, we investigated several methods to define a set of descriptors. The performance of each descriptor was tested on computer vision datasets for texture and scene categorization.

This work is ongoing and was presented as a poster at the 1st Workshop of Visual Image Interpretation in Humans and Machine (VIIHM, EPSRC Network for Biological and Computer Vision in the UK).

Shifting stimulus for faster receptive estimation of ensembles of neurons

Participants : Daniela Pamplona, Bruno Cessac, Pierre Kornprobst [correspondent] .

The spike triggered averaged (STA) technique has been widely used to estimate the receptive fields (RF) of sensory neurons  [58] . Theoretically, it has been shown that when the neurons are stimulated with a white noise stimulus the STA is an unbiased estimator of the neuron RF (up to a multiplicative constant). The error decreases with the number of spikes at a rate proportional to the stimulus variance  [75] . Experimentally, for visual neurons, the standard stimuli are checkerboards where block size is heuristically tuned. This raises difficulties when dealing with large neurons assemblies: When the block size is too small, neuron's response might be too weak, and when it is too large, one may miss RFs. Previously online updating the stimulus in the direction of larger stimulus-neural response correlation  [62] or mutual information  [69] , [68] has been proposed. However, these approaches can not be applied for an ensemble of cells recorded simultaneously since each neuron would update the stimulus in a different direction. We propose an improved checkerboard stimulus where blocks are shifted randomly in space at fixed time steps. Theoretically, we show that the STA remains an unbiased estimator of the RF. Additionally, we show two major properties of this new stimulus: (i) For a fixed block sized, RF spatial resolution is improved as a function of the number of possible shifts; (ii) Targeting a given RF spatial resolution, our method converges faster than the standard one. Numerically, we perform an exhaustive analysis of the performance of the approach based on simulated spiked trains from LNP cascades neurons varying RF sizes and positions. Results show global improvements in the RF representation even after short stimulation times. This makes this approach a promising solution to improve RF estimation of large ensemble of neurons.

This work is ongoing and was submitted to COSYNE 2015.

Shifting stimulus for faster receptive estimation of ensembles of neurons

Participants : Kartheek Medathati, Fabio Solari [University of Genoa - DIBRIS, Italy] , Manuela Chessa [University of Genoa - DIBRIS, Italy] , Guillaume S. Masson [Institut des Neurosciences de la Timone, Team InVibe] , Pierre Kornprobst [correspondent] .

Motion estimation has been studied extensively in neurosciences in the last two decades. The general consensus that has evolved from the studies in the primate vision is that it is done in a two stage process involving cortical areas V1 and MT in the brain. Spatio temporal filters are leading contenders in terms of models that capture the characteristics exhibited in these areas. Even though there are many models in the biological vision literature covering the optical flow estimation problem based on the spatio-temporal filters little is known in terms of their performance on the modern day computer vision datasets such as Middlebury. In this paper, we start from a mostly classical feedforward V1-MT model introducing a additional decoding step to obtain an optical flow estimation. Two extensions are also discussed using nonlinear filtering of the MT response for a better handling of discontinuities. One essential contribution of this paper is to show how a neural model can be adapted to deal with real sequences and it is here for the first time that such a neural model is benchmarked on the modern computer vision dataset Middlebury. Results are promising and suggest several possible improvements.

This work is ongoing and was presented as a poster at the 1st Workshop of Visual Image Interpretation in Humans and Machine (VIIHM, EPSRC Network for Biological and Computer Vision in the UK). See [35] for more details.

Exploring the richness of center-surround dynamics: A bifurcation study

Participants : Kartheek Medathati, James Rankin [Center for Neural Sciences, NYU, USA] , Guillaume S. Masson [Institut des Neurosciences de la Timone, Team InVibe] , Pierre Kornprobst [correspondent] .

The balance of excitatory and inhibitory interactions between neurons is one of the characteristic aspects of neural computation. In both neural network and neural field models these interactions have been modeled using center-surround connectivity kernels. Depending on the relative strength of excitation and inhibition these networks have been found to exhibit rich and interesting dynamical behavior. Although many models have been reported in the literature using center-surround connectivity kernels and many experimental studies have shown evidence for changes in observed behavior from winner-take-all to gain control, a thorough bifurcation analysis of these networks in terms of sensitivity of the network to peak strength, discriminability of the peaks and speed of convergence has not been done. In our present work we visit this question in order to identify the parameter regimes where this important switch in the behavior of the network occurs and also establish the trade offs that arise with the choice of a particular connectivity kernel.

This work is ongoing and was presented as a poster at the conference "Nonlinear dynamics and stochastic methods: from neuroscience to other biological applications"

From Habitat to Retina:Neural Population Coding using Natural Movies

Participants : Bruno Cessac [correspondent] , Ruben Herzog [Centro Interdisciplinario de Neurociencia de Valparaíso, Univ de Valparaíso, 2360102 Valparaíso, Chile] , Joaquin Araya [Centro Interdisciplinario de Neurociencia de Valparaíso, Univ de Valparaíso, 2360102 Valparaíso, Chile] , Michael Pizarro [Centro Interdisciplinario de Neurociencia de Valparaíso, Univ de Valparaíso, 2360102 Valparaíso, Chile] , Cesar Ravello [Centro Interdisciplinario de Neurociencia de Valparaíso, Univ de Valparaíso, 2360102 Valparaíso, Chile] , Maria Jose Escobar [Universidad Técnica Federico Santa María, Valparaiso, Chile] , Adrian Palacios [Centro Interdisciplinario de Neurociencia de Valparaíso, Univ de Valparaíso, 2360102 Valparaíso, Chile] .

We use a diurnal rodent retina (O. Degus), which has the advantage of present a 3:1 proportion of rods and cones, respectively, to study the RGC population responses to habitat-based natural stimuli. In order to do this, we have developed a mobile robot that is capable to record movies in the natural habitat of this rodent, simulating both his movements and the eye-ground distance, which allows us to stimulate and record an in vitro retina patch using MEA (multi electrode array) with a sequence of images taken from the animal natural habitat. The analysis of spike statistics has been done using the Enas software to characterize spatio-temporal pairwise correlation with Gibbs distributions. potential constitutes a useful tool for comparing pairwise spatio-temporal correlations between different conditions for the same RGC population. We show that correlated spiking patterns represents a major deviation between White Noise and Natural Movies conditions. We also conclude that population coding for this monophasic OFF RGC population is mostly based on spatial correlation when stimulated with Natural Movies.

This ongoing work has been presented in [48] .