A spiking neural network model of the medial superior olive using spike timing dependent plasticity for sound localization

Predicting the Influence of Axon Myelination on Sound Localization Precision Using a Spiking Neural Network Model of Auditory Brainstem

Spatial hearing allows animals to rapidly detect and localize auditory events in the surrounding environment. The auditory brainstem plays a central role in processing and extracting binaural spatial cues through microsecond-precise binaural integration, especially for detecting interaural time differences (ITDs) of low-frequency sounds at the medial superior olive (MSO). A series of mechanisms exist in the underlying neural circuits for preserving accurate action potential timing across multiple fibers, synapses and nuclei along this pathway. One of these is the myelination of afferent fibers that ensures reliable and temporally precise action potential propagation in the axon. There are several reports of fine-tuned myelination patterns in the MSO circuit, but how specifically myelination influences the precision of sound localization remains incompletely understood. Here we present a spiking neural network (SNN) model of the Mongolian gerbil auditory brainstem with myelinated axons to investigate whether different axon myelination thicknesses alter the sound localization process. Our model demonstrates that axon myelin thickness along the contralateral pathways can substantially modulate ITD detection. Furthermore, optimal ITD sensitivity is reached when the MSO receives contralateral inhibition via thicker myelinated axons compared to contralateral excitation, a result that is consistent with previously reported experimental observations. Our results suggest specific roles of axon myelination for extracting temporal dynamics in ITD decoding, especially in the pathway of the contralateral inhibition.

Axonal Conduction Delay Shapes the Precision of the Spatial Hearing in A Spiking Neural Network Model of Auditory Brainstem

One method by which the mammalian sound localization pathway localizes sound sources is by analyzing the microsecond-level difference between the arrival times of a sound at the two ears. However, how the neural circuits in the auditory brainstem precisely integrate signals from the two ears, and what the underlying mechanisms are, remains to be understood. Recent studies have reported that variations of axon myelination in the auditory brainstem produces various axonal conduction velocities and sophisticated temporal dynamics, which have not been well characterized in most existing models of sound localization circuits. Here, we present a spiking neural network model of the auditory brainstem to investigate how axon myelinations affect the precision of sound localization. Sound waves with different interaural time differences (ITDs) are encoded and used as stimuli, and the axon properties in the network are adjusted, and the corresponding axonal conduction delays are computed with a multi-compartment axon model. Through the simulation, the sensitivity of ITD perception varies with the myelin thickness of axons in the contralateral input pathways to the medial superior olive (MSO). The ITD perception becomes more precise when the contralateral inhibitory input propagates faster than the contralateral excitatory input. These results indicate that axon myelination and contralateral spike timing influence spatial hearing perception.

A Functional Spiking Neural Network of Ultra Compact Neurons

We demonstrate that recently introduced ultra-compact neurons (UCN) with a minimal number of components can be interconnected to implement a functional spiking neural network. For concreteness we focus on the Jeffress model, which is a classic neuro-computational model proposed in the 40’s to explain the sound directionality detection by animals and humans. In addition, we introduce a long-axon neuron, whose architecture is inspired by the Hodgkin-Huxley axon delay-line and where the UCNs implement the nodes of Ranvier. We then interconnect two of those neurons to an output layer of UCNs, which detect coincidences between spikes propagating down the long-axons. This functional spiking neural neuron circuit with biological relevance is built from identical UCN blocks, which are simple enough to be made with off-the-shelf electronic components. Our work realizes a new, accessible and affordable physical model platform, where neuroscientists can construct arbitrary mid-size spiking neuronal networks in a lego-block like fashion that work in continuous time. This should enable them to address in a novel experimental manner fundamental questions about the nature of the neural code and to test predictions from mathematical models and algorithms of basic neurobiology research. The present work aims at opening a new experimental field of basic research in Spiking Neural Networks to a potentially large community, which is at the crossroads of neurobiology, dynamical systems, theoretical neuroscience, condensed matter physics, neuromorphic engineering, artificial intelligence, and complex systems.

A Model for R(t) Elements and R(t) -Based Spike-Timing-Dependent Plasticity With Basic Circuit Examples

Spike-timing-dependent plasticity (STDP) is a fundamental synaptic learning rule observed in biology that leads to numerous behavioral and cognitive outcomes. Emulating STDP in electronic spiking neural networks with high-density memristive synapses is, therefore, of significant interest. While one popular method involves pulse-shaping the spiking neuron output voltages, an alternative approach is outlined in this article. The proposed STDP implementation uses time-varying dynamic resistance [ R ( t )] elements to achieve local synaptic learning from spike-pair STDP, spike triplet STDP, and firing rates. The R ( t ) elements are connected to each neuron circuit, thereby maintaining synaptic density and leveraging voltage division as a means of altering synaptic weight (memristor voltage). Example R ( t ) elements with their corresponding behaviors are demonstrated through simulation. A three-input-two-output network using single-memristor synaptic connections and R ( t ) elements is also simulated. Network-level effects, such as nonspecific synaptic plasticity, are discussed. Finally, spatiotemporal pattern recognition (STPR) using R ( t ) elements is demonstrated in simulation.

A Supervised Learning Algorithm for Learning Precise Timing of Multiple Spikes in Multilayer Spiking Neural Networks

There is a biological evidence to prove information is coded through precise timing of spikes in the brain. However, training a population of spiking neurons in a multilayer network to fire at multiple precise times remains a challenging task. Delay learning and the effect of a delay on weight learning in a spiking neural network (SNN) have not been investigated thoroughly. This paper proposes a novel biologically plausible supervised learning algorithm for learning precisely timed multiple spikes in a multilayer SNNs. Based on the spike-timing-dependent plasticity learning rule, the proposed learning method trains an SNN through the synergy between weight and delay learning. The weights of the hidden and output neurons are adjusted in parallel. The proposed learning method captures the contribution of synaptic delays to the learning of synaptic weights. Interaction between different layers of the network is realized through biofeedback signals sent by the output neurons. The trained SNN is used for the classification of spatiotemporal input patterns. The proposed learning method also trains the spiking network not to fire spikes at undesired times which contribute to misclassification. Experimental evaluation on benchmark data sets from the UCI machine learning repository shows that the proposed method has comparable results with classical rate-based methods such as deep belief network and the autoencoder models. Moreover, the proposed method can achieve higher classification accuracies than single layer and a similar multilayer SNN.

BrainFrame: a node-level heterogeneous accelerator platform for neuron simulations

OBJECTIVE

The advent of high-performance computing (HPC) in recent years has led to its increasing use in brain studies through computational models. The scale and complexity of such models are constantly increasing, leading to challenging computational requirements. Even though modern HPC platforms can often deal with such challenges, the vast diversity of the modeling field does not permit for a homogeneous acceleration platform to effectively address the complete array of modeling requirements.

APPROACH

In this paper we propose and build BrainFrame, a heterogeneous acceleration platform that incorporates three distinct acceleration technologies, an Intel Xeon-Phi CPU, a NVidia GP-GPU and a Maxeler Dataflow Engine. The PyNN software framework is also integrated into the platform. As a challenging proof of concept, we analyze the performance of BrainFrame on different experiment instances of a state-of-the-art neuron model, representing the inferior-olivary nucleus using a biophysically-meaningful, extended Hodgkin-Huxley representation. The model instances take into account not only the neuronal-network dimensions but also different network-connectivity densities, which can drastically affect the workload's performance characteristics.

MAIN RESULTS

The combined use of different HPC technologies demonstrates that BrainFrame is better able to cope with the modeling diversity encountered in realistic experiments while at the same time running on significantly lower energy budgets. Our performance analysis clearly shows that the model directly affects performance and all three technologies are required to cope with all the model use cases.

SIGNIFICANCE

The BrainFrame framework is designed to transparently configure and select the appropriate back-end accelerator technology for use per simulation run. The PyNN integration provides a familiar bridge to the vast number of models already available. Additionally, it gives a clear roadmap for extending the platform support beyond the proof of concept, with improved usability and directly useful features to the computational-neuroscience community, paving the way for wider adoption.

Extraction of Inter-Aural Time Differences Using a Spiking Neuron Network Model of the Medial Superior Olive

The mammalian auditory system is able to extract temporal and spectral features from sound signals at the two ears. One important cue for localization of low-frequency sound sources in the horizontal plane are inter-aural time differences (ITDs) which are first analyzed in the medial superior olive (MSO) in the brainstem. Neural recordings of ITD tuning curves at various stages along the auditory pathway suggest that ITDs in the mammalian brainstem are not represented in form of a Jeffress-type place code. An alternative is the hemispheric opponent-channel code, according to which ITDs are encoded as the difference in the responses of the MSO nuclei in the two hemispheres. In this study, we present a physiologically-plausible, spiking neuron network model of the mammalian MSO circuit and apply two different methods of extracting ITDs from arbitrary sound signals. The network model is driven by a functional model of the auditory periphery and physiological models of the cochlear nucleus and the MSO. Using a linear opponent-channel decoder, we show that the network is able to detect changes in ITD with a precision down to 10 μs and that the sensitivity of the decoder depends on the slope of the ITD-rate functions. A second approach uses an artificial neuronal network to predict ITDs directly from the spiking output of the MSO and ANF model. Using this predictor, we show that the MSO-network is able to reliably encode static and time-dependent ITDs over a large frequency range, also for complex signals like speech.

DL-ReSuMe: A Delay Learning-Based Remote Supervised Method for Spiking Neurons

Recent research has shown the potential capability of spiking neural networks (SNNs) to model complex information processing in the brain. There is biological evidence to prove the use of the precise timing of spikes for information coding. However, the exact learning mechanism in which the neuron is trained to fire at precise times remains an open problem. The majority of the existing learning methods for SNNs are based on weight adjustment. However, there is also biological evidence that the synaptic delay is not constant. In this paper, a learning method for spiking neurons, called delay learning remote supervised method (DL-ReSuMe), is proposed to merge the delay shift approach and ReSuMe-based weight adjustment to enhance the learning performance. DL-ReSuMe uses more biologically plausible properties, such as delay learning, and needs less weight adjustment than ReSuMe. Simulation results have shown that the proposed DL-ReSuMe approach achieves learning accuracy and learning speed improvements compared with ReSuMe.

Fast neuromorphic sound localization for binaural hearing aids

We report on the neuromorphic sound localization circuit which can enhance the perceptual sensation in a hearing aid system. All elements are simple leaky integrate-and-fire neuron circuits with different parameters optimized to suppress the impacts of synaptic circuit noises. The detection range and resolution of the proposed neuromorphic circuit are 500 us and 5 us, respectively. Our results show that, the proposed technique can localize a sound pulse with extremely narrow duration (∼ 1 ms) resulting in real-time response.

Spiking neural network model of sound localization using the interaural intensity difference

In this paper, a spiking neural network (SNN) architecture to simulate the sound localization ability of the mammalian auditory pathways using the interaural intensity difference cue is presented. The lateral superior olive was the inspiration for the architecture, which required the integration of an auditory periphery (cochlea) model and a model of the medial nucleus of the trapezoid body. The SNN uses leaky integrate-and-fire excitatory and inhibitory spiking neurons, facilitating synapses and receptive fields. Experimentally derived head-related transfer function (HRTF) acoustical data from adult domestic cats were employed to train and validate the localization ability of the architecture, training used the supervised learning algorithm called the remote supervision method to determine the azimuthal angles. The experimental results demonstrate that the architecture performs best when it is localizing high-frequency sound data in agreement with the biology, and also shows a high degree of robustness when the HRTF acoustical data is corrupted by noise.