Design and Analysis of a Neuromemristive Reservoir Computing Architecture for Biosignal Processing

Evolving Dual-Threshold Bienenstock-Cooper-Munro Learning Rules in Echo State Networks

The strengthening and the weakening of synaptic strength in existing Bienenstock-Cooper-Munro (BCM) learning rule are determined by a long-term potentiation (LTP) sliding modification threshold and the afferent synaptic activities. However, synaptic long-term depression (LTD) even affects low-active synapses during the induction of synaptic plasticity, which may lead to information loss. Biological experiments have found another LTD threshold that can induce either potentiation or depression or no change, even at the activated synapses. In addition, existing BCM learning rules can only select a set of fixed rule parameters, which is biologically implausible and practically inflexible to learn the structural information of input signals. In this article, an evolved dual-threshold BCM learning rule is proposed to regulate the reservoir internal connection weights of the echo-state-network (ESN), which can contribute to alleviating information loss and enhancing learning performance by introducing different optimal LTD thresholds for different postsynaptic neurons. Our experimental results show that the evolved dual-threshold BCM learning rule can result in the synergistic learning of different plasticity rules, effectively improving the learning performance of an ESN in comparison with existing neural plasticity learning rules and some state-of-the-art ESN variants on three widely used benchmark tasks and the prediction of an esterification process.

Seizure Detection and Prediction by Parallel Memristive Convolutional Neural Networks

During the past two decades, epileptic seizure detection and prediction algorithms have evolved rapidly. However, despite significant performance improvements, their hardware implementation using conventional technologies, such as Complementary Metal-Oxide-Semiconductor (CMOS), in power and area-constrained settings remains a challenging task; especially when many recording channels are used. In this paper, we propose a novel low-latency parallel Convolutional Neural Network (CNN) architecture that has between 2-2,800x fewer network parameters compared to State-Of-The-Art (SOTA) CNN architectures and achieves 5-fold cross validation accuracy of 99.84% for epileptic seizure detection, and 99.01% and 97.54% for epileptic seizure prediction, when evaluated using the University of Bonn Electroencephalogram (EEG), CHB-MIT and SWEC-ETHZ seizure datasets, respectively. We subsequently implement our network onto analog crossbar arrays comprising Resistive Random-Access Memory (RRAM) devices, and provide a comprehensive benchmark by simulating, laying out, and determining hardware requirements of the CNN component of our system. We parallelize the execution of convolution layer kernels on separate analog crossbars to enable 2 orders of magnitude reduction in latency compared to SOTA hybrid Memristive-CMOS Deep Learning (DL) accelerators. Furthermore, we investigate the effects of non-idealities on our system and investigate Quantization Aware Training (QAT) to mitigate the performance degradation due to low Analog-to-Digital Converter (ADC)/Digital-to-Analog Converter (DAC) resolution. Finally, we propose a stuck weight offsetting methodology to mitigate performance degradation due to stuck [Formula: see text] memristor weights, recovering up to 32% accuracy, without requiring retraining. The CNN component of our platform is estimated to consume approximately 2.791 W of power while occupying an area of 31.255 mm² in a 22 nm FDSOI CMOS process.

An Echo State Network Imparts a Curve Fitting

Recurrent neural networks (RNNs) are successfully employed in processing information from temporal data. Approaches to training such networks are varied and reservoir computing-based attainments, such as the echo state network (ESN), provide great ease in training. Akin to many machine learning algorithms rendering an interpolation function or fitting a curve, we observe that a driven system, such as an RNN, renders a continuous curve fitting if and only if it satisfies the echo state property. The domain of the learned curve is an abstract space of the left-infinite sequence of inputs and the codomain is the space of readout values. When the input originates from discrete-time dynamical systems, we find theoretical conditions under which a topological conjugacy between the input and reservoir dynamics can exist and present some numerical results relating the linearity in the reservoir to the forecasting abilities of the ESNs.

Echo State Network for Soft Actuator Control

Conventional model theories are not suitable to control soft-bodied robots as deformable materials present rapidly changing behaviors. Neuromorphic electronics are now entering the field of robotics, demonstrating that a highly integrated device can mimic the fundamental properties of a sensory synaptic system, including learning and proprioception. This research work focuses on the physical implementation of a reservoir computing-based network to actuate a soft-bodied robot. More specifically, modeling the hysteresis of a shape memory alloy (SMA) using echo state networks (ESN) in real-world situations represents a novel approach to enable soft machines with task-learning. In this work, we show that not only does our ESN model enable our SMA-based robot with locomotion, but it also discovers a successful strategy to do so. Compared to standard control modeling, established either by theoretical frameworks or from experimental data, here, we gained knowledge a posteriori, guided by the physical interactions between the trained model and the controlled actuator, interactions from which striking patterns emerged, and informed us about what type of locomotion would work best for our robot.

Opportunities for neuromorphic computing algorithms and applications

Neuromorphic computing technologies will be important for the future of computing, but much of the work in neuromorphic computing has focused on hardware development. Here, we review recent results in neuromorphic computing algorithms and applications. We highlight characteristics of neuromorphic computing technologies that make them attractive for the future of computing and we discuss opportunities for future development of algorithms and applications on these systems.

Spoken Digit Classification by In-Materio Reservoir Computing With Neuromorphic Atomic Switch Networks

Atomic Switch Networks comprising silver iodide (AgI) junctions, a material previously unexplored as functional memristive elements within highly interconnected nanowire networks, were employed as a neuromorphic substrate for physical Reservoir Computing This new class of ASN-based devices has been physically characterized and utilized to classify spoken digit audio data, demonstrating the utility of substrate-based device architectures where intrinsic material properties can be exploited to perform computation in-materio. This work demonstrates high accuracy in the classification of temporally analyzed Free-Spoken Digit Data These results expand upon the class of viable memristive materials available for the production of functional nanowire networks and bolster the utility of ASN-based devices as unique hardware platforms for neuromorphic computing applications involving memory, adaptation and learning.

Adaptive Extreme Edge Computing for Wearable Devices

Wearable devices are a fast-growing technology with impact on personal healthcare for both society and economy. Due to the widespread of sensors in pervasive and distributed networks, power consumption, processing speed, and system adaptation are vital in future smart wearable devices. The visioning and forecasting of how to bring computation to the edge in smart sensors have already begun, with an aspiration to provide adaptive extreme edge computing. Here, we provide a holistic view of hardware and theoretical solutions toward smart wearable devices that can provide guidance to research in this pervasive computing era. We propose various solutions for biologically plausible models for continual learning in neuromorphic computing technologies for wearable sensors. To envision this concept, we provide a systematic outline in which prospective low power and low latency scenarios of wearable sensors in neuromorphic platforms are expected. We successively describe vital potential landscapes of neuromorphic processors exploiting complementary metal-oxide semiconductors (CMOS) and emerging memory technologies (e.g., memristive devices). Furthermore, we evaluate the requirements for edge computing within wearable devices in terms of footprint, power consumption, latency, and data size. We additionally investigate the challenges beyond neuromorphic computing hardware, algorithms and devices that could impede enhancement of adaptive edge computing in smart wearable devices.

sEMG-Based Neural Network Prediction Model Selection of Gesture Fatigue and Dataset Optimization

The fatigue energy consumption of independent gestures can be obtained by calculating the power spectrum of surface electromyography (sEMG) signals. The existing research studies focus on the fatigue of independent gestures, while the research studies on integrated gestures are few. However, the actual gesture operation mode is usually integrated by multiple independent gestures, so the fatigue degree of integrated gestures can be predicted by training neural network of independent gestures. Three natural gestures including browsing information, playing games, and typing are divided into nine independent gestures in this paper, and the predicted model is established and trained by calculating the energy consumption of independent gestures. The artificial neural networks (ANNs) including backpropagation (BP) neural network, recurrent neural network (RNN), and long short-term memory (LSTM) are used to predict the fatigue of gesture. The support vector machine (SVM) is used to assist verification. Mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE) are utilized to evaluate the optimal prediction model. Furthermore, the different datasets of the processed sEMG signal and its decomposed wavelet coefficients are trained, respectively, and the changes of error functions of them are compared. The experimental results show that LSTM model is more suitable for gesture fatigue prediction. The processed sEMG signals are appropriate for using as the training set the fatigue degree of one-handed gesture. It is better to use wavelet decomposition coefficients as datasets to predict the high-dimensional sEMG signals of two-handed gestures. The experimental results can be applied to predict the fatigue degree of complex human-machine interactive gestures, help to avoid unreasonable gestures, and improve the user's interactive experience.

Stability and memory-loss go hand-in-hand: three results in dynamics and computation

The search for universal laws that help establish a relationship between dynamics and computation is driven by recent expansionist initiatives in biologically inspired computing. A general setting to understand both such dynamics and computation is a driven dynamical system that responds to a temporal input. Surprisingly, we find memory-loss a feature of driven systems to forget their internal states helps provide unambiguous answers to the following fundamental stability questions that have been unanswered for decades: what is necessary and sufficient so that slightly different inputs still lead to mostly similar responses? How does changing the driven system’s parameters affect stability? What is the mathematical definition of the edge-of-criticality? We anticipate our results to be timely in understanding and designing biologically inspired computers that are entering an era of dedicated hardware implementations for neuromorphic computing and state-of-the-art reservoir computing applications.

Recent advances in physical reservoir computing: A review

Reservoir computing is a computational framework suited for temporal/sequential data processing. It is derived from several recurrent neural network models, including echo state networks and liquid state machines. A reservoir computing system consists of a reservoir for mapping inputs into a high-dimensional space and a readout for pattern analysis from the high-dimensional states in the reservoir. The reservoir is fixed and only the readout is trained with a simple method such as linear regression and classification. Thus, the major advantage of reservoir computing compared to other recurrent neural networks is fast learning, resulting in low training cost. Another advantage is that the reservoir without adaptive updating is amenable to hardware implementation using a variety of physical systems, substrates, and devices. In fact, such physical reservoir computing has attracted increasing attention in diverse fields of research. The purpose of this review is to provide an overview of recent advances in physical reservoir computing by classifying them according to the type of the reservoir. We discuss the current issues and perspectives related to physical reservoir computing, in order to further expand its practical applications and develop next-generation machine learning systems.

Multiobjective Learning in the Model Space for Time Series Classification

A well-defined distance is critical for the performance of time series classification. Existing distance measurements can be categorized into two branches. One is to utilize handmade features for calculating distance, e.g., dynamic time warping, which is limited to exploiting the dynamic information of time series. The other methods make use of the dynamic information by approximating the time series with a generative model, e.g., Fisher kernel. However, previous distance measurements for time series seldom exploit the label information, which is helpful for classification by distance metric learning. In order to attain the benefits of the dynamic information of time series and the label information simultaneously, this paper proposes a multiobjective learning algorithm for both time series approximation and classification, termed multiobjective model-metric (MOMM) learning. In MOMM, a recurrent network is exploited as the temporal filter, based on which, a generative model is learned for each time series as a representation of that series. The models span a non-Euclidean space, where the label information is utilized to learn the distance metric. The distance between time series is then calculated as the model distance weighted by the learned metric. The network size is also optimized to learn parsimonious representations. MOMM simultaneously optimizes the data representation, the time series model separation, and the network size. The experiments show that MOMM achieves not only superior overall performance on uni/multivariate time series classification but also promising time series prediction performance.

Polymer Analog Memristive Synapse with Atomic-Scale Conductive Filament for Flexible Neuromorphic Computing System

With the advent of artificial intelligence (AI), memristors have received significant interest as a synaptic building block for neuromorphic systems, where each synaptic memristor should operate in an analog fashion, exhibiting multilevel accessible conductance states. Here, we demonstrate that the transition of the operation mode in poly(1,3,5-trivinyl-1,3,5-trimethyl cyclotrisiloxane) (pV3D3)-based flexible memristor from conventional binary to synaptic analog switching can be achieved simply by reducing the size of the formed filament. With the quantized conductance states observed in the flexible pV3D3 memristor, analog potentiation and depression characteristics of the memristive synapse are obtained through the growth of atomically thin Cu filament and lateral dissolution of the filament via dominant electric field effect, respectively. The face classification capability of our memristor is evaluated via simulation using an artificial neural network consisting of pV3D3 memristor synapses. These results will encourage the development of soft neuromorphic intelligent systems.

Reservoir computing using dynamic memristors for temporal information processing

Reservoir computing systems utilize dynamic reservoirs having short-term memory to project features from the temporal inputs into a high-dimensional feature space. A readout function layer can then effectively analyze the projected features for tasks, such as classification and time-series analysis. The system can efficiently compute complex and temporal data with low-training cost, since only the readout function needs to be trained. Here we experimentally implement a reservoir computing system using a dynamic memristor array. We show that the internal ionic dynamic processes of memristors allow the memristor-based reservoir to directly process information in the temporal domain, and demonstrate that even a small hardware system with only 88 memristors can already be used for tasks, such as handwritten digit recognition. The system is also used to experimentally solve a second-order nonlinear task, and can successfully predict the expected output without knowing the form of the original dynamic transfer function.

Reservoir computing facilitates the projection of temporal input signals onto a high-dimensional feature space via a dynamic system, known as the reservoir. Du et al. realise this concept using metal-oxide-based memristors with short-term memory to perform digit recognition tasks and solve non-linear problems.