Bat echolocation processing using first-spike latency coding

First-spike coding promotes accurate and efficient spiking neural networks for discrete events with rich temporal structures

Spiking neural networks (SNNs) are well-suited to process asynchronous event-based data. Most of the existing SNNs use rate-coding schemes that focus on firing rate (FR), and so they generally ignore the spike timing in events. On the contrary, methods based on temporal coding, particularly time-to-first-spike (TTFS) coding, can be accurate and efficient but they are difficult to train. Currently, there is limited research on applying TTFS coding to real events, since traditional TTFS-based methods impose one-spike constraint, which is not realistic for event-based data. In this study, we present a novel decision-making strategy based on first-spike (FS) coding that encodes FS timings of the output neurons to investigate the role of the first-spike timing in classifying real-world event sequences with complex temporal structures. To achieve FS coding, we propose a novel surrogate gradient learning method for discrete spike trains. In the forward pass, output spikes are encoded into discrete times to generate FS times. In the backpropagation, we develop an error assignment method that propagates error from FS times to spikes through a Gaussian window, and then supervised learning for spikes is implemented through a surrogate gradient approach. Additional strategies are introduced to facilitate the training of FS timings, such as adding empty sequences and employing different parameters for different layers. We make a comprehensive comparison between FS and FR coding in the experiments. Our results show that FS coding achieves comparable accuracy to FR coding while leading to superior energy efficiency and distinct neuronal dynamics on data sequences with very rich temporal structures. Additionally, a longer time delay in the first spike leads to higher accuracy, indicating important information is encoded in the timing of the first spike.

Advancements in Algorithms and Neuromorphic Hardware for Spiking Neural Networks

Artificial neural networks (ANNs) have experienced a rapid advancement for their success in various application domains, including autonomous driving and drone vision. Researchers have been improving the performance efficiency and computational requirement of ANNs inspired by the mechanisms of the biological brain. Spiking neural networks (SNNs) provide a power-efficient and brain-inspired computing paradigm for machine learning applications. However, evaluating large-scale SNNs on classical von Neumann architectures (central processing units/graphics processing units) demands a high amount of power and time. Therefore, hardware designers have developed neuromorphic platforms to execute SNNs in and approach that combines fast processing and low power consumption. Recently, field-programmable gate arrays (FPGAs) have been considered promising candidates for implementing neuromorphic solutions due to their varied advantages, such as higher flexibility, shorter design, and excellent stability. This review aims to describe recent advances in SNNs and the neuromorphic hardware platforms (digital, analog, hybrid, and FPGA based) suitable for their implementation. We present that biological background of SNN learning, such as neuron models and information encoding techniques, followed by a categorization of SNN training. In addition, we describe state-of-the-art SNN simulators. Furthermore, we review and present FPGA-based hardware implementation of SNNs. Finally, we discuss some future directions for research in this field.

FNS allows efficient event-driven spiking neural network simulations based on a neuron model supporting spike latency

Neural modelling tools are increasingly employed to describe, explain, and predict the human brain's behavior. Among them, spiking neural networks (SNNs) make possible the simulation of neural activity at the level of single neurons, but their use is often threatened by the resources needed in terms of processing capabilities and memory. Emerging applications where a low energy burden is required (e.g. implanted neuroprostheses) motivate the exploration of new strategies able to capture the relevant principles of neuronal dynamics in reduced and efficient models. The recent Leaky Integrate-and-Fire with Latency (LIFL) spiking neuron model shows some realistic neuronal features and efficiency at the same time, a combination of characteristics that may result appealing for SNN-based brain modelling. In this paper we introduce FNS, the first LIFL-based SNN framework, which combines spiking/synaptic modelling with the event-driven approach, allowing us to define heterogeneous neuron groups and multi-scale connectivity, with delayed connections and plastic synapses. FNS allows multi-thread, precise simulations, integrating a novel parallelization strategy and a mechanism of periodic dumping. We evaluate the performance of FNS in terms of simulation time and used memory, and compare it with those obtained with neuronal models having a similar neurocomputational profile, implemented in NEST, showing that FNS performs better in both scenarios. FNS can be advantageously used to explore the interaction within and between populations of spiking neurons, even for long time-scales and with a limited hardware configuration.

Active head rolls enhance sonar-based auditory localization performance

Animals utilize a variety of active sensing mechanisms to perceive the world around them. Echolocating bats are an excellent model for the study of active auditory localization. The big brown bat (Eptesicus fuscus), for instance, employs active head roll movements during sonar prey tracking. The function of head rolls in sound source localization is not well understood. Here, we propose an echolocation model with multi-axis head rotation to investigate the effect of active head roll movements on sound localization performance. The model autonomously learns to align the bat's head direction towards the target. We show that a model with active head roll movements better localizes targets than a model without head rolls. Furthermore, we demonstrate that active head rolls also reduce the time required for localization in elevation. Finally, our model offers key insights to sound localization cues used by echolocating bats employing active head movements during echolocation.

Temporal coding of echo spectral shape in the bat auditory cortex

Echolocating bats rely upon spectral interference patterns in echoes to reconstruct fine details of a reflecting object’s shape. However, the acoustic modulations required to do this are extremely brief, raising questions about how their auditory cortex encodes and processes such rapid and fine spectrotemporal details. Here, we tested the hypothesis that biosonar target shape representation in the primary auditory cortex (A1) is more reliably encoded by changes in spike timing (latency) than spike rates and that latency is sufficiently precise to support a synchronization-based ensemble representation of this critical auditory object feature space. To test this, we measured how the spatiotemporal activation patterns of A1 changed when naturalistic spectral notches were inserted into echo mimic stimuli. Neurons tuned to notch frequencies were predicted to exhibit longer latencies and lower mean firing rates due to lower signal amplitudes at their preferred frequencies, and both were found to occur. Comparative analyses confirmed that significantly more information was recoverable from changes in spike times relative to concurrent changes in spike rates. With this data, we reconstructed spatiotemporal activation maps of A1 and estimated the level of emerging neuronal spike synchrony between cortical neurons tuned to different frequencies. The results support existing computational models, indicating that spectral interference patterns may be efficiently encoded by a cascading tonotopic sequence of neural synchronization patterns within an ensemble of network activity that relates to the physical features of the reflecting object surface.

Echolocating bats rely upon spectral interference patterns in echoes to reconstruct fine details of a reflecting object’s shape. This study shows that the latency shifts induced by spectral notch patterns can provide the foundation for an avalanche of neuronal synchrony that is sufficient to support encoding of auditory object shape features during active biosonar.

A Neuro-Inspired System for Online Learning and Recognition of Parallel Spike Trains, Based on Spike Latency, and Heterosynaptic STDP

Humans perform remarkably well in many cognitive tasks including pattern recognition. However, the neuronal mechanisms underlying this process are not well understood. Nevertheless, artificial neural networks, inspired in brain circuits, have been designed and used to tackle spatio-temporal pattern recognition tasks. In this paper we present a multi-neuronal spike pattern detection structure able to autonomously implement online learning and recognition of parallel spike sequences (i.e., sequences of pulses belonging to different neurons/neural ensembles). The operating principle of this structure is based on two spiking/synaptic neurocomputational characteristics: spike latency, which enables neurons to fire spikes with a certain delay and heterosynaptic plasticity, which allows the own regulation of synaptic weights. From the perspective of the information representation, the structure allows mapping a spatio-temporal stimulus into a multi-dimensional, temporal, feature space. In this space, the parameter coordinate and the time at which a neuron fires represent one specific feature. In this sense, each feature can be considered to span a single temporal axis. We applied our proposed scheme to experimental data obtained from a motor-inhibitory cognitive task. The results show that out method exhibits similar performance compared with other classification methods, indicating the effectiveness of our approach. In addition, its simplicity and low computational cost suggest a large scale implementation for real time recognition applications in several areas, such as brain computer interface, personal biometrics authentication, or early detection of diseases.

Latency modulation of collicular neurons induced by electric stimulation of the auditory cortex in Hipposideros pratti: In vivo intracellular recording

In the auditory pathway, the inferior colliculus (IC) receives and integrates excitatory and inhibitory inputs from the lower auditory nuclei, contralateral IC, and auditory cortex (AC), and then uploads these inputs to the thalamus and cortex. Meanwhile, the AC modulates the sound signal processing of IC neurons, including their latency (i.e., first-spike latency). Excitatory and inhibitory corticofugal projections to the IC may shorten and prolong the latency of IC neurons, respectively. However, the synaptic mechanisms underlying the corticofugal latency modulation of IC neurons remain unclear. Thus, this study probed these mechanisms via in vivo intracellular recording and acoustic and focal electric stimulation. The AC latency modulation of IC neurons is possibly mediated by pre-spike depolarization duration, pre-spike hyperpolarization duration, and spike onset time. This study suggests an effective strategy for the timing sequence determination of auditory information uploaded to the thalamus and cortex.

Brian hears: online auditory processing using vectorization over channels

The human cochlea includes about 3000 inner hair cells which filter sounds at frequencies between 20 Hz and 20 kHz. This massively parallel frequency analysis is reflected in models of auditory processing, which are often based on banks of filters. However, existing implementations do not exploit this parallelism. Here we propose algorithms to simulate these models by vectorizing computation over frequency channels, which are implemented in “Brian Hears,” a library for the spiking neural network simulator package “Brian.” This approach allows us to use high-level programming languages such as Python, because with vectorized operations, the computational cost of interpretation represents a small fraction of the total cost. This makes it possible to define and simulate complex models in a simple way, while all previous implementations were model-specific. In addition, we show that these algorithms can be naturally parallelized using graphics processing units, yielding substantial speed improvements. We demonstrate these algorithms with several state-of-the-art cochlear models, and show that they compare favorably with existing, less flexible, implementations.