Csaba Petre

Emergence of bottom-up saliency in a spiking model of V1

We present anatomically detailed spiking model of the parvo and magno pathways of the retina, primary visual cortex (V1), and superior colliculus (SC) to enable active saccades. Due to STDP and visual experience, the model shows the emergence of a saliency map, resulting in the perceptual behavior of bottom-up (pop-out) attention. In contrast to previous models proposed to explain pop-out based attention for simple features (e.g., saliency map hypothesis of [1]), where feature selectivity and inhibitory mechanisms between similar features are pre-wired, connectivity in our spiking model is neither pre-wired nor are neurons pre-labeled, but feature selectivity and pop-out behavior still emerges. Projections between cell types in the V1 model (L4 and L2/3) are in agreement with anatomical data. Both excitatory and inhibitory synapses are subject to different forms of STDP. These plasticity mechanisms and exposure to rich natural visual stimuli lead to (i) neuronal responses similar to those recorded in vivo, (ii - parvo) formation in color selective cells, and (iii - magno) formation of simple and complex cells covering a broad range of orientations and spatial frequencies. Pop-out mechanism is mediated by modifying the activity in layer 2/3 with long-range effective inhibition using a narrow form of STDP, which selectively picks up short temporal correlations between neurons responding to similar features but depresses ignores neurons responding to different features. Stronger within-feature long-range inhibition dampens the population response to features that are abundant in the input, but allows strong response to salient input features. The activity of V1 drives the SC, resulting in pop-out saccades. (The SC model is presented in a separate submission.) The model connects electrophysiology (spiking activity) and perception, and it explains animal behavior in a variety of standard pop-out tasks. Neurons responding to vertical features will receive strong inhibition from other vertical neurons, therefore weakening their response, while response triggered by the single horizontal bar remains strong. Figure 1 A. Input image; B. V1 layer 2/3 activity without long-range inhibition; C. V1 layer 2/3 activity with long-range inhibition; D. Superior colliculus activity that directly drives the saccadic mechanism. Activity is averaged over two seconds.

Latency and rate coding in a spiking model of retina

We describe a family of spiking retina computational models for large-scale simulations of the visual pathways. The models reproduce the overall spatial, temporal, and chromatic structure of the receptive fields of midget and parasol retinal ganglion cells (RGCs) of the primate retina, as well as their contrast response, while using fairly simple models of bipolar, horizontal, amacrine, and RGCs. These retina models provide input to a realistic model of visual cortex, presented in a separate submission. The retina model simulates cone and bipolar cell dynamics using a damped wave equation. Parameters are so chosen that an impulse retinal input gives rise to a damped biphasic output current, and that a coherent stimulus elicits a substantially stronger response than an incoherent stimulus of the same power (Figure ​(Figure1).1). Static or low-frequency visual inputs are smoothly rejected. In contrast to biology, this stage of the model is linear; rectification and adaptation mechanisms are all realized in the spiking ganglion cell layer. In the ganglion cell layer parasol, midget, and SBC cells are simulated. The spatio-temporal receptive fields of different RGCs are not prewired; they emerge via the lateral interactions between the horizontal, bipolar, and amacrine cells. Spike generation in the RGCs is modeled using spiking neurons with two dynamic variables and an intrinsic adaptation mechanism. Midget RGCs have an additional leaky temporal integration of the input current, to account for the relatively tonic character of their responses. Parasol RGCs have an additional partial divisive normalization of the input current, to reproduce the saturating parasol response to relative contrast of the stimulus. Simplified amacrine-cell dynamics is introduced to provide for the spatiotemporal structure of the parasol receptive field surrounds. Figure 1 A-E: model response to coherent vs. randomized stimulus. Moving edge stimulus is presented (A) and elicits strong response (B). Same stimulus with fixed random permutation of pixels (C) elicits much weaker response (D). (E) compares the total responses ... Desired latency encoding of input features is achieved by calibrating the parameters of the spiking RGC conductance dynamics. A near logarithmic latency to the first spike, for a step up in the input signal, provides for a good contrast invariance of the relative spike timing in the response. A combination of spike-latency and spike-count encoding of the input stimulus, together with the enhanced response of the simulated retina to coherent stimuli, results in an informative yet not too noisy spiking input to the LGN and V1; the information content analysis of the RGC spike-train vocabulary will be presented. This retina model was used to produce a fully-emergent orientation tuning in a spiking model [1] of V1 cortex.

Beyond inhibition: lateral modulation of plasticity of feedforward synapses in a spiking model of V1

Lateral inhibition is typically used to repel neural receptive fields. Here we introduce an additional learning mechanism that modifies the plasticity for feedforward synapses based on lateral interactions. We show a model, based on evidence from biological recordings[1], in which a heterosynaptic learning rule in conjunction with lateral inhibition introduces competition among neurons for input features. We demonstrate an STDP learning rule for feedforward connections to a spiking neuron where plasticity is modulated by activity of neighboring neurons. We apply the learning rule to a spiking model of primary visual cortex. In our model, input to V1 cortical neurons comes from the magnocellular pathway of a simulated spiking retina responding to saccades over a natural image. A group of neurons in V1 respond to a particular input feature and convey information that they have spiked to neighboring neurons by way of specialized lateral connections. The connections convey a direct signal of recent spiking activity. Alternatively, lateral inhibitory synapses and the level of inhibition to a neuron can also be used as this signal. If this recent neighbor activity signal is high, feedforward plasticity to the neuron becomes a simple flat depression as opposed to regular exponential STDP, as shown in Figure ​Figure1A.1A. Thus, late spiking neurons are prevented from developing receptive fields similar to those of earlier spiking neurons for a given input feature. This rule in conjunction with lateral inhibition and slow weight updrift ensures competition between neurons for features over a long timescale. Figure 1 A. Detail of the heterosynaptic rule. Neurons N1 and N2 fire early for the input, and their feedforward synapses undergo normal STDP. They send an activity signal to N3 and N4, whose feedforward synaptic weights are then depressed to prevent learning ... Our V1 model is built on a spatially distributed grid of single-compartment spiking neurons with parameters configured to model regular spiking cortical pyramidal cells, as detailed in [2]. The model has plastic excitatory feedforward connections and local lateral inhibition between spatially proximal neurons. In our model, we observe a better overall coverage of orientation selectivity of V1 Layer 4 neurons with our heterosynaptic rule than without it. We find a larger range of orientations and less redundancy of receptive field features between neighboring neurons, as shown in Figure ​Figure1B.1B. Such a rule could be used in a generic spiking cortical architecture to enforce independence of neural receptive fields.

Balanced excitation and inhibition in a spiking model of V1

Experimental studies have shown that neuronal excitation is balanced with inhibition and spikes are triggered only when that fine balance is perturbed. It is also known that inhibition is critical for receptive field tuning, yet it is not clear what role is played by different types of inhibitory interneurons and how the corresponding balanced circuitry could emerge via spike timing dependent plasticity (STDP). To study these questions we have constructed a large-scale detailed spiking model of V1 involving a variety of simulated neurons: fast-spiking (FS) interneurons, low threshold spiking (LTS) interneurons and regular spiking (RS) neurons. We modeled layer 4 and layer 2/3 of the primary visual cortex and a number of projections between cell types in agreement with anatomical data. Synaptic dynamics is governed by a set of STDP and activity dependent plasticity mechanisms for both inhibitory and excitatory synapses. The plasticity rules have been chosen to be in quantitative agreement with experiment where the data is available. For many of connections however, the data is either unavailable or noisy. In these cases plasticity rules were chosen based on a guided guess constrained by the requirement of structural stability of the system and expected response properties of cells to probing stimuli. Together, the plasticity rules lead to stable neuronal response and formation of orientation-selective receptive fields. The network learns simple and complex cells of a broad range of orientations and spatial frequencies. The model converges to a balanced neurodynamics and biologically reasonable firing rates. Our study shows that in the presence of strong thalamic drive, plastic inhibition is necessary for feature selectivity. The FS cells remove DC component of the input while firing of the LTS cells imposes sparse response and balances out feedback excitation.