Jens Kremkow

PyNN: A Common Interface for Neuronal Network Simulators

Computational neuroscience has produced a diversity of software for simulations of networks of spiking neurons, with both negative and positive consequences. On the one hand, each simulator uses its own programming or configuration language, leading to considerable difficulty in porting models from one simulator to another. This impedes communication between investigators and makes it harder to reproduce and build on the work of others. On the other hand, simulation results can be cross-checked between different simulators, giving greater confidence in their correctness, and each simulator has different optimizations, so the most appropriate simulator can be chosen for a given modelling task. A common programming interface to multiple simulators would reduce or eliminate the problems of simulator diversity while retaining the benefits. PyNN is such an interface, making it possible to write a simulation script once, using the Python programming language, and run it without modification on any supported simulator (currently NEURON, NEST, PCSIM, Brian and the Heidelberg VLSI neuromorphic hardware). PyNN increases the productivity of neuronal network modelling by providing high-level abstraction, by promoting code sharing and reuse, and by providing a foundation for simulator-agnostic analysis, visualization and data-management tools. PyNN increases the reliability of modelling studies by making it much easier to check results on multiple simulators. PyNN is open-source software and is available from http://neuralensemble.org/PyNN.

A comprehensive workflow for general-purpose neural modeling with highly configurable neuromorphic hardware systems

In this article, we present a methodological framework that meets novel requirements emerging from upcoming types of accelerated and highly configurable neuromorphic hardware systems. We describe in detail a device with 45 million programmable and dynamic synapses that is currently under development, and we sketch the conceptual challenges that arise from taking this platform into operation. More specifically, we aim at the establishment of this neuromorphic system as a flexible and neuroscientifically valuable modeling tool that can be used by non-hardware experts. We consider various functional aspects to be crucial for this purpose, and we introduce a consistent workflow with detailed descriptions of all involved modules that implement the suggested steps: The integration of the hardware interface into the simulator-independent model description language PyNN; a fully automated translation between the PyNN domain and appropriate hardware configurations; an executable specification of the future neuromorphic system that can be seamlessly integrated into this biology-to-hardware mapping process as a test bench for all software layers and possible hardware design modifications; an evaluation scheme that deploys models from a dedicated benchmark library, compares the results generated by virtual or prototype hardware devices with reference software simulations and analyzes the differences. The integration of these components into one hardware–software workflow provides an ecosystem for ongoing preparative studies that support the hardware design process and represents the basis for the maturity of the model-to-hardware mapping software. The functionality and flexibility of the latter is proven with a variety of experimental results.

Gating of Signal Propagation in Spiking Neural Networks by Balanced and Correlated Excitation and Inhibition

Both ongoing and natural stimulus driven neuronal activity are dominated by transients. Selective gating of these transients is mandatory for proper brain function and may, in fact, form the basis of millisecond-fast decision making and action selection. Here we propose that neuronal networks may exploit timing differences between correlated excitation and inhibition (temporal gating) to control the propagation of spiking activity transients. When combined with excitation–inhibition balance, temporal gating constitutes a powerful mechanism to control the propagation of mixtures of transient and tonic neural activity components.

Functional consequences of correlated excitatory and inhibitory conductances in cortical networks

Neurons in the neocortex receive a large number of excitatory and inhibitory synaptic inputs. Excitation and inhibition dynamically balance each other, with inhibition lagging excitation by only few milliseconds. To characterize the functional consequences of such correlated excitation and inhibition, we studied models in which this correlation structure is induced by feedforward inhibition (FFI). Simple circuits show that an effective FFI changes the integrative behavior of neurons such that only synchronous inputs can elicit spikes, causing the responses to be sparse and precise. Further, effective FFI increases the selectivity for propagation of synchrony through a feedforward network, thereby increasing the stability to background activity. Last, we show that recurrent random networks with effective inhibition are more likely to exhibit dynamical network activity states as have been observed in vivo. Thus, when a feedforward signal path is embedded in such recurrent network, the stabilizing effect of effective inhibition creates an suitable substrate for signal propagation. In conclusion, correlated excitation and inhibition support the notion that synchronous spiking may be important for cortical processing.

Neuronal nonlinearity explains greater visual spatial resolution for darks than lights

Significance Light and dark stimuli are separately processed by ON and OFF channels in retina and thalamus. Although most textbooks assume that ON and OFF visual responses are relatively balanced throughout the visual system, recent studies have identified a pronounced overrepresentation of OFF responses in the cerebral cortex. This recent discovery resonates with Galileo and Helmholtz’s pioneering observations that visual spatial resolution is higher for darks than lights. In this paper, we demonstrate that these two seemingly separate findings are related and caused by a pronounced difference between ON and OFF luminance response functions, which most likely originates in photoreceptors. Therefore, asymmetric ON and OFF neural responses provide the neurophysiological explanation for an almost four-century-old puzzle dating back to Galileo. Astronomers and physicists noticed centuries ago that visual spatial resolution is higher for dark than light stimuli, but the neuronal mechanisms for this perceptual asymmetry remain unknown. Here we demonstrate that the asymmetry is caused by a neuronal nonlinearity in the early visual pathway. We show that neurons driven by darks (OFF neurons) increase their responses roughly linearly with luminance decrements, independent of the background luminance. However, neurons driven by lights (ON neurons) saturate their responses with small increases in luminance and need bright backgrounds to approach the linearity of OFF neurons. We show that, as a consequence of this difference in linearity, receptive fields are larger in ON than OFF thalamic neurons, and cortical neurons are more strongly driven by darks than lights at low spatial frequencies. This ON/OFF asymmetry in linearity could be demonstrated in the visual cortex of cats, monkeys, and humans and in the cat visual thalamus. Furthermore, in the cat visual thalamus, we show that the neuronal nonlinearity is present at the ON receptive field center of ON-center neurons and ON receptive field surround of OFF-center neurons, suggesting an origin at the level of the photoreceptor. These results demonstrate a fundamental difference in visual processing between ON and OFF channels and reveal a competitive advantage for OFF neurons over ON neurons at low spatial frequencies, which could be important during cortical development when retinal images are blurred by immature optics in infant eyes.

Principles underlying sensory map topography in primary visual cortex

The primary visual cortex contains a detailed map of the visual scene, which is represented according to multiple stimulus dimensions including spatial location, ocular dominance and stimulus orientation. The maps for spatial location and ocular dominance arise from the spatial arrangement of thalamic afferent axons in the cortex. However, the origins of the other maps remain unclear. Here we show that the cortical maps for orientation, direction and retinal disparity in the cat (Felis catus) are all strongly related to the organization of the map for spatial location of light (ON) and dark (OFF) stimuli, an organization that we show is OFF-dominated, OFF-centric and runs orthogonal to ocular dominance columns. Because this ON–OFF organization originates from the clustering of ON and OFF thalamic afferents in the visual cortex, we conclude that all main features of visual cortical topography, including orientation, direction and retinal disparity, follow a common organizing principle that arranges thalamic axons with similar retinotopy and ON–OFF polarity in neighbouring cortical regions.

Neuronal and perceptual differences in the temporal processing of darks and lights

Visual information is mediated by two major thalamic pathways that signal light decrements (OFF) and increments (ON) in visual scenes, the OFF pathway being faster than the ON. Here, we demonstrate that this OFF temporal advantage is transferred to visual cortex and has a correlate in human perception. OFF-dominated cortical neurons in cats responded ∼3 ms faster to visual stimuli than ON-dominated cortical neurons, and dark-mediated suppression in ON-dominated neurons peaked ∼14 ms faster than light-mediated suppression in OFF-dominated neurons. Consistent with the neuronal differences, human observers were 6-14 ms faster at detecting darks than lights and better at discriminating dark than light flickers. Neuronal and perceptual differences both vanished if backgrounds were biased toward darks. Our results suggest that the cortical OFF pathway is faster than the ON pathway at increasing and suppressing visual responses, and these differences have parallels in the human visual perception of lights and darks.

In Vivo Monosynaptic Excitatory Transmission between Layer 2 Cortical Pyramidal Neurons

Summary Little is known about the properties of monosynaptic connections between identified neurons in vivo. We made multiple (two to four) two-photon targeted whole-cell recordings from neighboring layer 2 mouse somatosensory barrel cortex pyramidal neurons in vivo to investigate excitatory monosynaptic transmission in the hyperpolarized downstate. We report that pyramidal neurons form a sparsely connected (6.7% connectivity) network with an overrepresentation of bidirectional connections. The majority of unitary excitatory postsynaptic potentials were small in amplitude (<0.5 mV), with a small minority >1 mV. The coefficient of variation (CV = 0.74) could largely be explained by the presence of synaptic failures (22%). Both the CV and failure rates were reduced with increasing amplitude. The mean paired-pulse ratio was 1.15 and positively correlated with the CV. Our approach will help bridge the gap between connectivity and function and allow investigations into the impact of brain state on monosynaptic transmission and integration.

RESPONSE PROPERTIES OF LOCAL FIELD POTENTIALS AND NEIGHBORING SINGLE NEURONS IN AWAKE PRIMARY VISUAL CORTEX

Recordings from local field potentials (LFPs) are becoming increasingly common in research and clinical applications, but we still have a poor understanding of how LFP stimulus selectivity originates from the combined activity of single neurons. Here, we systematically compared the stimulus selectivity of LFP and neighboring single-unit activity (SUA) recorded in area primary visual cortex (V1) of awake primates. We demonstrate that LFP and SUA have similar stimulus preferences for orientation, direction of motion, contrast, size, temporal frequency, and even spatial phase. However, the average SUA had 50 times better signal-to-noise, 20% higher contrast sensitivity, 45% higher direction selectivity, and 15% more tuning depth than the average LFP. Low LFP frequencies (<30 Hz) were most strongly correlated with the spiking frequencies of neurons with nonlinear spatial summation and poor orientation/direction selectivity that were located near cortical current sinks (negative LFPs). In contrast, LFP gamma frequencies (>30 Hz) were correlated with a more diverse group of neurons located near cortical sources (positive LFPs). In summary, our results indicate that low- and high-frequency LFP pool signals from V1 neurons with similar stimulus preferences but different response properties and cortical depths.

Columnar organization of spatial phase in visual cortex

Images are processed in the primary visual cortex by neurons that encode different stimulus orientations and spatial phases. In primates and carnivores, neighboring cortical neurons share similar orientation preferences, but spatial phases were thought to be randomly distributed. We discovered a columnar organization for spatial phase in cats that shares similarities with the columnar organization for orientation. For both orientation and phase, the mean difference across vertically aligned neurons was less than one-fourth of a cycle. Cortical neurons showed threefold more diversity in phase than orientation preference; however, the average phase of local neuronal populations was similar through the depth of layer 4. We conclude that columnar organization for visual space is not only defined by the spatial location of the stimulus, but also by absolute phase. Taken together with previous findings, our results suggest that this phase-visuotopy is responsible for the emergence of orientation maps.