A spiking neural network model of self-organized pattern recognition in the early mammalian olfactory system

From Brain Models to Robotic Embodied Cognition: How Does Biological Plausibility Inform Neuromorphic Systems?

We examine the challenging "marriage" between computational efficiency and biological plausibility-A crucial node in the domain of spiking neural networks at the intersection of neuroscience, artificial intelligence, and robotics. Through a transdisciplinary review, we retrace the historical and most recent constraining influences that these parallel fields have exerted on descriptive analysis of the brain, construction of predictive brain models, and ultimately, the embodiment of neural networks in an enacted robotic agent. We study models of Spiking Neural Networks (SNN) as the central means enabling autonomous and intelligent behaviors in biological systems. We then provide a critical comparison of the available hardware and software to emulate SNNs for investigating biological entities and their application on artificial systems. Neuromorphics is identified as a promising tool to embody SNNs in real physical systems and different neuromorphic chips are compared. The concepts required for describing SNNs are dissected and contextualized in the new no man's land between cognitive neuroscience and artificial intelligence. Although there are recent reviews on the application of neuromorphic computing in various modules of the guidance, navigation, and control of robotic systems, the focus of this paper is more on closing the cognition loop in SNN-embodied robotics. We argue that biologically viable spiking neuronal models used for electroencephalogram signals are excellent candidates for furthering our knowledge of the explainability of SNNs. We complete our survey by reviewing different robotic modules that can benefit from neuromorphic hardware, e.g., perception (with a focus on vision), localization, and cognition. We conclude that the tradeoff between symbolic computational power and biological plausibility of hardware can be best addressed by neuromorphics, whose presence in neurorobotics provides an accountable empirical testbench for investigating synthetic and natural embodied cognition. We argue this is where both theoretical and empirical future work should converge in multidisciplinary efforts involving neuroscience, artificial intelligence, and robotics.

A stochastic model for interacting neurons in the olfactory bulb

We focus on interacting neurons organized in a block-layered network devoted to the information processing from the sensory system to the brain. Specifically, we consider the firing activity of olfactory sensory neurons, periglomerular, granule and mitral cells in the context of the neuronal activity of the olfactory bulb. We propose and investigate a stochastic model of a layered and modular network to describe the dynamic behavior of each prototypical neuron, taking into account both its role (excitatory/inhibitory) and its location within the network. We adopt specific Gauss-Markov processes suitable to provide reliable estimates of the firing activity of the different neurons, given their linkages. Furthermore, we study the impact of selective excitation/inhibition on the information transmission by means of simulations and numerical estimates obtained through a Volterra integral approach.

An Investigation into Spike-Based Neuromorphic Approaches for Artificial Olfactory Systems

The implementation of neuromorphic methods has delivered promising results for vision and auditory sensors. These methods focus on mimicking the neuro-biological architecture to generate and process spike-based information with minimal power consumption. With increasing interest in developing low-power and robust chemical sensors, the application of neuromorphic engineering concepts for electronic noses has provided an impetus for research focusing on improving these instruments. While conventional e-noses apply computationally expensive and power-consuming data-processing strategies, neuromorphic olfactory sensors implement the biological olfaction principles found in humans and insects to simplify the handling of multivariate sensory data by generating and processing spike-based information. Over the last decade, research on neuromorphic olfaction has established the capability of these sensors to tackle problems that plague the current e-nose implementations such as drift, response time, portability, power consumption and size. This article brings together the key contributions in neuromorphic olfaction and identifies future research directions to develop near-real-time olfactory sensors that can be implemented for a range of applications such as biosecurity and environmental monitoring. Furthermore, we aim to expose the computational parallels between neuromorphic olfaction and gustation for future research focusing on the correlation of these senses.

A Principle for Describing and Verifying Brain Mechanisms Using Ongoing Activity

Not even the most informed scientist can setup a theory that takes all brain signals into account. A neuron not only receives neuronal short range and long range input from all over the brain but a neuron also receives input from the extracellular space, astrocytes and vasculature. Given this complexity, how does one describe and verify a typical brain mechanism in vivo? Common to most described mechanisms is that one focuses on how one specific input signal gives rise to the activity in a population of neurons. This can be an input from a brain area, a population of neurons or a specific cell type. All remaining inputs originating from all over the brain are lumped together into one background input. The division into two inputs is attractive since it can be used to quantify the relative importance of either input. Here we have chosen to extract the specific and the background input by means of recording and inhibiting the specific input. We summarize what it takes to estimate the two inputs on a single trial level. The inhibition should not only be strong but also fast and the specific input measurement has to be tailor-made to the inhibition. In essence, we suggest ways to control electrophysiological experiments in vivo. By applying those controls it may become possible to describe and verify many brain mechanisms, and it may also allow the study of the integration of spontaneous and ongoing activity, which in turn governs cognition and behavior.

A model of signal processing at the isolated hair cell of the frog semicircular canal

A computational model has been developed to simulate the electrical behavior of the type II hair cell dissected from the crista ampullaris of frog semicircular canals. In its basolateral membrane, it hosts a system of four voltage-dependent conductances (g _A , g _KV , g _KCa , g _Ca ). The conductance behavior was mathematically described using original patch-clamp experimental data. The transient K current, IA, was isolated as the difference between the currents obtained before and after removing IA inactivation. The remaining current, IKD, results from the summation of a voltage-dependent K current, IKV, a voltage-calcium-dependent K current, IKCa, and the calcium current, ICa. IKD was modeled as a single lumped current, since the physiological role of each component is actually not discernible. To gain a clear understanding of its prominent role in sustaining transmitter release at the cytoneural junction, ICa was modeled under different experimental conditions. The model includes the description of voltage- and time-dependent kinetics for each single current. After imposing any starting holding potential, the system sets the pertinent values of the variables and continually updates them in response to variations in membrane potential. The model reconstructs the individual I-V curves obtained in voltage-clamp experiments and simulations compare favorably with the experimental data. The model proves useful in describing the early steps of signal processing that results from the interaction of the apical receptor current with the basolateral voltage-dependent conductances. The program is thus helpful in understanding aspects of sensory transduction that are hard to analyze in the native hair cell of the crista ampullaris.

Functional Relevance of Different Basal Ganglia Pathways Investigated in a Spiking Model with Reward Dependent Plasticity

The brain enables animals to behaviorally adapt in order to survive in a complex and dynamic environment, but how reward-oriented behaviors are achieved and computed by its underlying neural circuitry is an open question. To address this concern, we have developed a spiking model of the basal ganglia (BG) that learns to dis-inhibit the action leading to a reward despite ongoing changes in the reward schedule. The architecture of the network features the two pathways commonly described in BG, the direct (denoted D1) and the indirect (denoted D2) pathway, as well as a loop involving striatum and the dopaminergic system. The activity of these dopaminergic neurons conveys the reward prediction error (RPE), which determines the magnitude of synaptic plasticity within the different pathways. All plastic connections implement a versatile four-factor learning rule derived from Bayesian inference that depends upon pre- and post-synaptic activity, receptor type, and dopamine level. Synaptic weight updates occur in the D1 or D2 pathways depending on the sign of the RPE, and an efference copy informs upstream nuclei about the action selected. We demonstrate successful performance of the system in a multiple-choice learning task with a transiently changing reward schedule. We simulate lesioning of the various pathways and show that a condition without the D2 pathway fares worse than one without D1. Additionally, we simulate the degeneration observed in Parkinson's disease (PD) by decreasing the number of dopaminergic neurons during learning. The results suggest that the D1 pathway impairment in PD might have been overlooked. Furthermore, an analysis of the alterations in the synaptic weights shows that using the absolute reward value instead of the RPE leads to a larger change in D1.

Large-Scale Simulations of Plastic Neural Networks on Neuromorphic Hardware

SpiNNaker is a digital, neuromorphic architecture designed for simulating large-scale spiking neural networks at speeds close to biological real-time. Rather than using bespoke analog or digital hardware, the basic computational unit of a SpiNNaker system is a general-purpose ARM processor, allowing it to be programmed to simulate a wide variety of neuron and synapse models. This flexibility is particularly valuable in the study of biological plasticity phenomena. A recently proposed learning rule based on the Bayesian Confidence Propagation Neural Network (BCPNN) paradigm offers a generic framework for modeling the interaction of different plasticity mechanisms using spiking neurons. However, it can be computationally expensive to simulate large networks with BCPNN learning since it requires multiple state variables for each synapse, each of which needs to be updated every simulation time-step. We discuss the trade-offs in efficiency and accuracy involved in developing an event-based BCPNN implementation for SpiNNaker based on an analytical solution to the BCPNN equations, and detail the steps taken to fit this within the limited computational and memory resources of the SpiNNaker architecture. We demonstrate this learning rule by learning temporal sequences of neural activity within a recurrent attractor network which we simulate at scales of up to 2.0 × 104 neurons and 5.1 × 107 plastic synapses: the largest plastic neural network ever to be simulated on neuromorphic hardware. We also run a comparable simulation on a Cray XC-30 supercomputer system and find that, if it is to match the run-time of our SpiNNaker simulation, the super computer system uses approximately 45× more power. This suggests that cheaper, more power efficient neuromorphic systems are becoming useful discovery tools in the study of plasticity in large-scale brain models.

Long-range recruitment of Martinotti cells causes surround suppression and promotes saliency in an attractor network model

Although the importance of long-range connections for cortical information processing has been acknowledged for a long time, most studies focused on the long-range interactions between excitatory cortical neurons. Inhibitory interneurons play an important role in cortical computation and have thus far been studied mainly with respect to their local synaptic interactions within the cortical microcircuitry. A recent study showed that long-range excitatory connections onto Martinotti cells (MC) mediate surround suppression. Here we have extended our previously reported attractor network of pyramidal cells (PC) and MC by introducing long-range connections targeting MC. We have demonstrated how the network with Martinotti cell-mediated long-range inhibition gives rise to surround suppression and also promotes saliency of locations at which simple non-uniformities in the stimulus field are introduced. Furthermore, our analysis suggests that the presynaptic dynamics of MC is only ancillary to its orientation tuning property in enabling the network with saliency detection. Lastly, we have also implemented a disinhibitory pathway mediated by another interneuron type (VIP interneurons), which inhibits MC and abolishes surround suppression.

Cortical Feedback Decorrelates Olfactory Bulb Output in Awake Mice

The olfactory bulb receives rich glutamatergic projections from the piriform cortex. However, the dynamics and importance of these feedback signals remain unknown. Here, we use multiphoton calcium imaging to monitor cortical feedback in the olfactory bulb of awake mice and further probe its impact on the bulb output. Responses of feedback boutons were sparse, odor specific, and often outlasted stimuli by several seconds. Odor presentation either enhanced or suppressed the activity of boutons. However, any given bouton responded with stereotypic polarity across multiple odors, preferring either enhancement or suppression. Feedback representations were locally diverse and differed in dynamics across bulb layers. Inactivation of piriform cortex increased odor responsiveness and pairwise similarity of mitral cells but had little impact on tufted cells. We propose that cortical feedback differentially impacts these two output channels of the bulb by specifically decorrelating mitral cell responses to enable odor separation.

Reducing the computational footprint for real-time BCPNN learning

The implementation of synaptic plasticity in neural simulation or neuromorphic hardware is usually very resource-intensive, often requiring a compromise between efficiency and flexibility. A versatile, but computationally-expensive plasticity mechanism is provided by the Bayesian Confidence Propagation Neural Network (BCPNN) paradigm. Building upon Bayesian statistics, and having clear links to biological plasticity processes, the BCPNN learning rule has been applied in many fields, ranging from data classification, associative memory, reward-based learning, probabilistic inference to cortical attractor memory networks. In the spike-based version of this learning rule the pre-, postsynaptic and coincident activity is traced in three low-pass-filtering stages, requiring a total of eight state variables, whose dynamics are typically simulated with the fixed step size Euler method. We derive analytic solutions allowing an efficient event-driven implementation of this learning rule. Further speedup is achieved by first rewriting the model which reduces the number of basic arithmetic operations per update to one half, and second by using look-up tables for the frequently calculated exponential decay. Ultimately, in a typical use case, the simulation using our approach is more than one order of magnitude faster than with the fixed step size Euler method. Aiming for a small memory footprint per BCPNN synapse, we also evaluate the use of fixed-point numbers for the state variables, and assess the number of bits required to achieve same or better accuracy than with the conventional explicit Euler method. All of this will allow a real-time simulation of a reduced cortex model based on BCPNN in high performance computing. More important, with the analytic solution at hand and due to the reduced memory bandwidth, the learning rule can be efficiently implemented in dedicated or existing digital neuromorphic hardware.

Electrical responses of three classes of granule cells of the olfactory bulb to synaptic inputs in different dendritic locations

This work consists of a computational study of the electrical responses of three classes of granule cells of the olfactory bulb to synaptic activation in different dendritic locations. The constructed models were based on morphologically detailed compartmental reconstructions of three granule cell classes of the olfactory bulb with active dendrites described by Bhalla and Bower (1993, pp. 1948-1965) and dendritic spine distributions described by Woolf et al. (1991, pp. 1837-1854). The computational studies with the model neurons showed that different quantities of spines have to be activated in each dendritic region to induce an action potential, which always was originated in the active terminal dendrites, independently of the location of the stimuli, and the morphology of the dendritic tree. These model predictions might have important computational implications in the context of olfactory bulb circuits.