Application Of Spiking Neural Networks Research Articles

Application of Neuromorphic Olfactory Approach for High-Accuracy Classification of Malts.

Current developments in artificial olfactory systems, also known as electronic nose (e-nose) systems, have benefited from advanced machine learning techniques that have significantly improved the conditioning and processing of multivariate feature-rich sensor data. These advancements are complemented by the application of bioinspired algorithms and architectures based on findings from neurophysiological studies focusing on the biological olfactory pathway. The application of spiking neural networks (SNNs), and concepts from neuromorphic engineering in general, are one of the key factors that has led to the design and development of efficient bioinspired e-nose systems. However, only a limited number of studies have focused on deploying these models on a natively event-driven hardware platform that exploits the benefits of neuromorphic implementation, such as ultra-low-power consumption and real-time processing, for simplified integration in a portable e-nose system. In this paper, we extend our previously reported neuromorphic encoding and classification approach to a real-world dataset that consists of sensor responses from a commercial e-nose system when exposed to eight different types of malts. We show that the proposed SNN-based classifier was able to deliver 97% accurate classification results at a maximum latency of 0.4 ms per inference with a power consumption of less than 1 mW when deployed on neuromorphic hardware. One of the key advantages of the proposed neuromorphic architecture is that the entire functionality, including pre-processing, event encoding, and classification, can be mapped on the neuromorphic system-on-a-chip (NSoC) to develop power-efficient and highly-accurate real-time e-nose systems.

ConvSNN: A surrogate gradient spiking neural framework for radar gesture recognition

<h2>Abstract</h2> Spiking neural networks (SNNs) have recently gained large interest for edge-AI applications due to their low latency and ultra-low energy consumption. Unlike DNNs, SNNs communicate information using spike trains. As the derivative of spike trains are highly ill-defined, the use of surrogate gradients has been proposed as an efficient method for training SNNs. Still, the lack of open-source SNN softwares and the limited range of demonstrated SNN applications slows down a wider SNN adoption. We release our ConvSNN framework, demonstrating the novel applicability of quantized-weight SNNs for radar gesture recognition. Our framework will facilitate future research in the SNN area.

Flexible On-Line Reconfiguration of Multi-Core Neuromorphic Platforms

Neuromorphic architectures are emerging not only for real-time simulation of brain-scale biological neural networks but also to support innovative brain-inspired computational paradigms. In both domains there is an increasing demand for flexibility in terms of network configuration and runtime redesign of network parameters and simulated neurons models. Due to the intrinsically high parallelism of these architectures and complexity of the interconnect, broadcasting updates to the cores is time consuming. Hence, static solutions where the network is reloaded from an external host instead of being reconfigured are highly inefficient. To address these requirements, we designed an Application Command Protocol (ACP). The proposed protocol provides a mechanism to remotely trigger the execution of high-level op-codes by the cores and manage their application memory, and supports a more flexible computational model and memory management. We worked on SpiNNaker, a multi-core globally-asynchronous locally-synchronous platform running Spiking Neural Networks (SNNs) simulations. We demonstrated ACP in two SNN applications: i) SNN configuration, where simulation data are efficiently generated through ACP in the memory of computing nodes and ii) SNN reconfiguration, where ACP is used to change SNN network parameters at runtime and to easily switch from learning to test phase in a SNN classification application.

Automated Adaptive Threshold-Based Feature Extraction and Learning for Spiking Neural Networks

Over the past years Spiking Neural Networks (SNNs) models became attractive as a possible bridge to enable low-power event-driven neuromorphic hardware. SNNs have a high computational power due to the implicit employment of the biologically inspired input times. SNNs employ various parameters such as neuron threshold, synaptic delays, and weights in their structures. However, SNNs applications are still limited and elementary compared with other neural network architectures such as the Convolution Neural Networks (CNNs). In this research, a new SNN-based model named Adaptive Threshold Module (ATM) and its algorithm are proposed. The proposed ATM and algorithm depend on the adaptation of the internal spiking neuron threshold level. Adapting the threshold of the neurons is employed to control the spiking neuron firing rate to uniquely extract the main features of the input pattern that is in the shape of spike trains. It is shown that this technique works as an automated feature extraction method of input patterns in an efficient and faster way than other methods. The proposed method can preserve all information of the input spike trains. Simulations of the proposed model and the algorithm, using the challenging speech TIDIGITS dataset, sound RWCP dataset, and Poisson distribution spike trains, show encouraging results. The ATM can make SNN provide an accuracy surpassing that of the current state-of-the-art SNN algorithms and conventional non-spiking learning models.

Electric Field-Induced Metal-Insulator Phase Separation in Lithium Titanates: Tuning Symmetry By Lithiation

Specialized hardware for neural networks requires materials with tunable symmetry, retention and speed at low power consumption. The vast majority of memristor are based on two types of ions: either oxygen vacancy migration, in the so-called Valence Change Memories (VCM), or a metal cation, usually Ag+ and Cu2+, in the so-called Electrochemical Metallization Cells (ECM). Despite their excellent performance, their widespread implementation in today’s integrated circuits is delayed due to the need to address cycle-to-cycle and device-to-device variabilities while circumventing electroforming and asymmetry, which are inherent issues associated to the filamentary nature of the switching mechanism. Recently, Li-ion is emerging as an alternative, given the higher diffusivity of Li+ when compared to oxygen, and the ability of Li-oxides solid state conductors to accumulate and deplete lithium at the interfaces and bulk. We have recently proposed lithium titanates, originally developed as Li-ion battery anode materials, as promising candidates for memristive-based neuromorphic computing hardware[1]. In this presentation, we will discuss the non-volatile, non-filamentary bipolar resistive switching characteristics of lithium titanates compounds, Li4+3xTi5O12, as a function of the lithiation degree. We have employed a recently proposed strategy to overcome lithium loss during thin film deposition and finely control the final lithiation degree of the films[2] to create stoichiometrically lithiated Li4Ti5O12 spinel phase and a highly lithiated Li7Ti5O12 rock- salt phase memristive devices. By using ex- and in-operando spectroscopy to monitor the Lithium filling and emptying of structural positions during electrochemical measurements, we investigate the controlled formation of a metallic phase (Li7Ti5O12) percolating through an insulating medium (Li4Ti5O12) with no volume changes under voltage bias, thereby controlling the spatially averaged conductivity of the film device. We present a theoretical model to explain the observed hysteretic switching behavior based on electrochemical nonequilibrium thermodynamics, in which the metal-insulator transition results from electrically driven phase separation of Li4Ti5O12 and Li7Ti5O12. Permittivity enhancement drives lithium ions to regions of high electric field intensity, which become metallic filaments above a critical applied bias, and the ions relax back to their low-conductivity initial state at lower voltages. One of the most striking outcomes is that the metal-insulator transition of llithium titanate can be uniquely modulated for neuromorphic computing purposes, such as control of the neural pulse train symmetry in conductance and the resistance on-to-off ratio, simply by adjusting the lithium stoichiometry and phase pattern of the films. We report ability of highly lithiated phase of Li7Ti5O12 for Deep Neural Network applications, given the large retentions and symmetry, and opportunity for the low lithiated phase of Li4Ti5O12 towards Spiking Neural Network applications, due to the shorter retention and large resistance changes. Our findings pave the way for lithium oxides to enable thin-film memristive devices with adjustable symmetry and retention.

Pattern Recognition of Spiking Neural Networks Based on Visual Mechanism and Supervised Synaptic Learning.

Electrophysiological studies have shown that mammalian primary visual cortex are selective for the orientations of visual stimuli. Inspired by this mechanism, we propose a hierarchical spiking neural network (SNN) for image classification. Grayscale input images are fed through a feed-forward network consisting of orientation-selective neurons, which then projected to a layer of downstream classifier neurons through the spiking-based supervised tempotron learning rule. Based on the orientation-selective mechanism of the visual cortex and tempotron learning rule, the network can effectively classify images of the extensively studied MNIST database of handwritten digits, which achieves 96% classification accuracy based on only 2000 training samples (traditional training set is 60000). Compared with other classification methods, our model not only guarantees the biological plausibility and the accuracy of image classification but also significantly reduces the needed training samples. Considering the fact that the most commonly used deep learning neural networks need big data samples and high power consumption in image recognition, this brain-inspired computational neural network model based on the layer-by-layer hierarchical image processing mechanism of the visual cortex may provide a basis for the wide application of spiking neural networks in the field of intelligent computing.

FusionSense: Emotion Classification Using Feature Fusion of Multimodal Data and Deep Learning in a Brain-Inspired Spiking Neural Network.

Using multimodal signals to solve the problem of emotion recognition is one of the emerging trends in affective computing. Several studies have utilized state of the art deep learning methods and combined physiological signals, such as the electrocardiogram (EEG), electroencephalogram (ECG), skin temperature, along with facial expressions, voice, posture to name a few, in order to classify emotions. Spiking neural networks (SNNs) represent the third generation of neural networks and employ biologically plausible models of neurons. SNNs have been shown to handle Spatio-temporal data, which is essentially the nature of the data encountered in emotion recognition problem, in an efficient manner. In this work, for the first time, we propose the application of SNNs in order to solve the emotion recognition problem with the multimodal dataset. Specifically, we use the NeuCube framework, which employs an evolving SNN architecture to classify emotional valence and evaluate the performance of our approach on the MAHNOB-HCI dataset. The multimodal data used in our work consists of facial expressions along with physiological signals such as ECG, skin temperature, skin conductance, respiration signal, mouth length, and pupil size. We perform classification under the Leave-One-Subject-Out (LOSO) cross-validation mode. Our results show that the proposed approach achieves an accuracy of 73.15% for classifying binary valence when applying feature-level fusion, which is comparable to other deep learning methods. We achieve this accuracy even without using EEG, which other deep learning methods have relied on to achieve this level of accuracy. In conclusion, we have demonstrated that the SNN can be successfully used for solving the emotion recognition problem with multimodal data and also provide directions for future research utilizing SNN for Affective computing. In addition to the good accuracy, the SNN recognition system is requires incrementally trainable on new data in an adaptive way. It only one pass training, which makes it suitable for practical and on-line applications. These features are not manifested in other methods for this problem.

Normalized approximate descent used for spike based automatic bird species recognition system

The paper involves a study on the application of spiking neural network (SNN), which is also known as the third generation of neural networks for automatic bird species recognition. Spiking neural network has worked incredibly well in Neuromorphic computing, such as sequence identification, character identification, etc. This neural network architecture is a consistent model, which comprises of sensory encrypt, training and decrypt functional parts. This decrypt layer utilizes an algorithm based on the normalized approximate gradient descent (NormAD) for synaptic weight adjustment. The result reflected in this paper is for processing spatiotemporal pattern, and has higher computational power over perceptron, along with biological plausibility of spiking neurons. The proposed representation uses segmentation of audio frame based on the energy of the bird call separating voiced, unvoiced and a silence portion of it. This is passed to the attribute extrication method i.e. Mel frequency Cepstral coefficient and the by-product is thus fed to the SNN system for bird species classification. The research output recognizes 14 bird species with 90.10% accuracy in comparison to the perceptron system with an accuracy of 78%.

Deep SCNN-Based Real-Time Object Detection for Self-Driving Vehicles Using LiDAR Temporal Data

Real-time accurate detection of three-dimensional (3D) objects is a fundamental necessity for self-driving vehicles. Most existing computer vision approaches are based on convolutional neural networks (CNNs). Although the CNN-based approaches can achieve high detection accuracy, their high energy consumption is a severe drawback. To resolve this problem, novel energy efficient approaches should be explored. Spiking neural network (SNN) is a promising candidate because it has orders-of-magnitude lower energy consumption than CNN. Unfortunately, the studying of SNN has been limited in small networks only. The application of SNN for large 3D object detection networks has remain largely open. In this paper, we integrate spiking convolutional neural network (SCNN) with temporal coding into the YOLOv2 architecture for real-time object detection. To take the advantage of spiking signals, we develop a novel data preprocessing layer that translates 3D point-cloud data into spike time data. We propose an analog circuit to implement the non-leaky integrate and fire neuron used in our SCNN, from which the energy consumption of each spike is estimated. Moreover, we present a method to calculate the network sparsity and the energy consumption of the overall network. Extensive experiments have been conducted based on the KITTI dataset, which show that the proposed network can reach competitive detection accuracy as existing approaches, yet with much lower average energy consumption. If implemented in dedicated hardware, our network could have a mean sparsity of 56.24% and extremely low total energy consumption of 0.247mJ only. Implemented in NVIDIA GTX 1080i GPU, we can achieve 35.7 fps frame rate, high enough for real-time object detection.

Application of spiking neural networks for modelling the process of high-temperature hydrogen production in systems with gas-cooled reactors *

Hydrogen energy is able to solve the problem of the dependence of modern industries on fossil fuels and significantly reduce the amount of harmful emissions. One of the ways to produce hydrogen is high-temperature water-steam electrolysis. Increasing the temperature of the steam involved in electrolysis makes the process more efficient. The key problem is the use of a reliable heat energy source capable of reaching high temperatures. High-temperature gas-cooled reactors with a gaseous coolant and a graphite moderator provide a solution to the problem of heating the electrolyte. Part of the heat energy is used for producing electrical energy required for electrolysis. Modern electrolyzers built as arrays of tubular or planar electrolytic cells with a nuclear energy source make it possible to produce hydrogen by decomposing water molecules, and the working temperature control leads to a decrease in the Nernst potential. The operation of such facilities is complicated by the need to determine the optimal parameters of the electrolysis cell, the steam flow rate, and the operating current density. To reduce the costs associated with the process optimization, it is proposed to use a low-temperature electrolysis system controlled by a spiking neural network. The results confirm the effectiveness of intelligent technologies that implement adaptive control of hybrid modeling processes in order to organize the most feasible hydrogen production in a specific process, the parameters of which can be modified depending on the specific use of the reactor thermal energy. In addition, the results of the study confirm the feasibility of using a combined functional structure made on the basis of spiking neurons to correct the parameters of the developed electrolytic system. The proposed simulation strategy can significantly reduce the consumption of computational resources in comparison with models based only on neural network prediction methods.

Применение спайковой нейронной сети для моделирования процесса высокотемпературного производства водорода в системах с газоохлаждаемыми реакторами

Применение спайковой нейронной сети для моделирования процесса высокотемпературного производства водорода в системах с газоохлаждаемыми реакторами

System-Level Simulation of Hardware Spiking Neural Network Based on Synaptic Transistors and I&amp;F Neuron Circuits

We perform a system-level simulation of hardware spiking neural network (SNN) consisting of silicon-based synaptic transistors and integrate-and-fire (I&F) neuron circuits. Using electrical models of the synaptic device and I&F neuron circuit, a three-layer fully connected SNN in hardware is presented for MNIST pattern recognition by means of ex situ training. Right-justified rate coding is employed as an information encoding method, and negative weight values are implemented by a pair of the synaptic transistors (specifically, excitatory and inhibitory synapses). Furthermore, the variability effect occurring in the devices and circuits is demonstrated. This result indicates that the system has tolerance to the variations and how precisely the variations need to be controlled for hardware SNN applications.

An Event-Driven Classifier for Spiking Neural Networks Fed with Synthetic or Dynamic Vision Sensor Data.

This paper introduces a novel methodology for training an event-driven classifier within a Spiking Neural Network (SNN) System capable of yielding good classification results when using both synthetic input data and real data captured from Dynamic Vision Sensor (DVS) chips. The proposed supervised method uses the spiking activity provided by an arbitrary topology of prior SNN layers to build histograms and train the classifier in the frame domain using the stochastic gradient descent algorithm. In addition, this approach can cope with leaky integrate-and-fire neuron models within the SNN, a desirable feature for real-world SNN applications, where neural activation must fade away after some time in the absence of inputs. Consequently, this way of building histograms captures the dynamics of spikes immediately before the classifier. We tested our method on the MNIST data set using different synthetic encodings and real DVS sensory data sets such as N-MNIST, MNIST-DVS, and Poker-DVS using the same network topology and feature maps. We demonstrate the effectiveness of our approach by achieving the highest classification accuracy reported on the N-MNIST (97.77%) and Poker-DVS (100%) real DVS data sets to date with a spiking convolutional network. Moreover, by using the proposed method we were able to retrain the output layer of a previously reported spiking neural network and increase its performance by 2%, suggesting that the proposed classifier can be used as the output layer in works where features are extracted using unsupervised spike-based learning methods. In addition, we also analyze SNN performance figures such as total event activity and network latencies, which are relevant for eventual hardware implementations. In summary, the paper aggregates unsupervised-trained SNNs with a supervised-trained SNN classifier, combining and applying them to heterogeneous sets of benchmarks, both synthetic and from real DVS chips.

Rapid application prototyping for hardware modular spiking neural network architectures

Spiking neural networks (SNNs) are well suited for functions such as data/pattern classification, estimation, prediction, signal processing and robotic control applications. Whereas the real-world embedded applications are often multi-functional with orthogonal or contradicting functional requirements. The EMBRACE hardware modular SNN architecture has been previously reported as an embedded computing platform for complex real-world applications. The EMBRACE architecture employs genetic algorithm (GA) for training the SNN which offers faster prototyping of SNN applications, but exhibits a number of limitations including poor scalability and search space explosions for the evolution of large-scale, complex, real-world applications. This paper investigates the limitations of evolving real-world embedded applications with orthogonal functional goals on hardware SNN using GA-based training. This paper presents a novel, fast and efficient application prototyping technique using the EMBRACE hardware modular SNN architecture and the GA-based evolution platform. Modular design and evolution of a robotic navigational controller application decomposed into obstacle avoidance controller and speed and direction manager application subtasks is presented. The proposed modular evolution technique successfully integrates the orthogonal functionalities of the application and helps to overcome contradicting application scenarios gracefully. Results illustrate that the modular evolution of the application reduces the SNN configuration search space and complexity for the GA-based SNN evolution, offering rapid and successful prototyping of complex applications on the hardware SNN platform. The paper presents validation results of the evolved robotic application implemented on the EMBRACE architecture prototyped on Xilinx Virtex-6 FPGA interacting with the player-stage robotics simulator.

Modelling Spiking Neural Network from the Architecture Evaluation Perspective

The brain-inspired spiking neural network (SNN) computing paradigm offers the potential for low-power and scalable computing, suited to many intelligent tasks that conventional computational systems find difficult. On the other hand, NoC (network-on-chips) based very large scale integration (VLSI) systems have been widely used to mimic neurobiological architectures (including SNNs). This paper proposes an evaluation methodology for SNN applications from the aspect of micro-architecture. First, we extract accurate SNN models from existing simulators of neural systems. Second, a cycle-accurate NoC simulator is implemented to execute the aforementioned SNN applications to get timing and energyconsumption information. We believe this method not only benefits the exploration of NoC design space but also bridges the gap between applications (especially those from the neuroscientists’ community) and neuromorphic hardware. Based on the method, we have evaluated some typical SNNs in terms of timing and energy. The method is valuable for the development of neuromorphic hardware and applications.

Fixed latency on-chip interconnect for hardware spiking neural network architectures

Information in a Spiking Neural Network (SNN) is encoded as the relative timing between spikes. Distortion in spike timings can impact the accuracy of SNN operation by modifying the precise firing time of neurons within the SNN. Maintaining the integrity of spike timings is crucial for reliable operation of SNN applications. A packet switched Network on Chip (NoC) infrastructure offers scalable connectivity for spike communication in hardware SNN architectures. However, shared resources in NoC architectures can result in unwanted variation in spike packet transfer latency. This packet latency jitter distorts the timing information conveyed on the synaptic connections in the SNN, resulting in unreliable application behaviour. This paper presents a SystemC simulation based analysis of the synaptic information distortion in NoC based hardware SNNs. The paper proposes a fixed spike transfer latency ring topology interconnect for spike communication between neural tiles, using a novel timestamped spike broadcast flow control scheme. The proposed architectural technique is evaluated using spike rates employed in previously reported mesh topology NoC based hardware SNN applications, which exhibited spike latency jitter over NoC paths. Results indicate that the proposed interconnect offers fixed spike transfer latency and eliminates the associated information distortion. The paper presents the micro-architecture of the proposed ring router. The FPGA validated ring interconnect architecture has been synthesised using 65nm low-power CMOS technology. Silicon area comparisons for various ring sizes are presented. Scalability of the proposed architecture has been addressed by employing a hierarchical NoC architecture.

Hardware spiking neural network prototyping and application

EMBRACE has been proposed as a scalable, reconfigurable, mixed signal, embedded hardware Spiking Neural Network (SNN) device. EMBRACE, which is yet to be realised, targets the issues of area, power and scalability through the use of a low area, low power analogue neuron/synapse cell, and a digital packet-based Network on Chip (NoC) communication architecture. The paper describes the implementation and testing of EMBRACE-FPGA, an FPGA-based hardware SNN prototype. The operation of the NoC inter-neuron communication approach and its ability to support large scale, reconfigurable, highly interconnected SNNs is illustrated. The paper describes an integrated training and configuration platform and an on-chip fitness function, which supports GA-based evolution of SNN parameters. The practicalities of using the SNN development platform and SNN configuration toolset are described. The paper considers the impact of latency jitter noise introduced by the NoC router and the EMBRACE-FPGA processor-based neuron/synapse model on SNN accuracy and evolution time. Benchmark SNN applications are described and results demonstrate the evolution of high quality and robust solutions in the presence of noise. The reconfigurable EMBRACE architecture enables future investigation of adaptive hardware applications and self repair in evolvable hardware.

Advances in Design and Application of Spiking Neural Networks

This paper presents new findings in the design and application of biologically plausible neural networks based on spiking neuron models, which represent a more plausible model of real biological neurons where time is considered as an important feature for information encoding and processing in the brain. The design approach consists of an evolutionary strategy based supervised training algorithm, newly developed by the authors, and the use of different biologically plausible neuronal models. A dynamic synapse (DS) based neuron model, a biologically more detailed model, and the spike response model (SRM) are investigated in order to demonstrate the efficacy of the proposed approach and to further our understanding of the computing capabilities of the nervous system. Unlike the conventional synapse, represented as a static entity with a fixed weight, employed in conventional and SRM-based neural networks, a DS is weightless and its strength changes upon the arrival of incoming input spikes. Therefore its efficacy depends on the temporal structure of the impinging spike trains. In the proposed approach, the training of the network free parameters is achieved using an evolutionary strategy where, instead of binary encoding, real values are used to encode the static and DS parameters which underlie the learning process. The results show that spiking neural networks based on both types of synapse are capable of learning non-linearly separable data by means of spatio-temporal encoding. Furthermore, a comparison of the obtained performance with classical neural networks (multi-layer perceptrons) is presented.