In natural language processing (NLP), the "Transformer" architecture was proposed as first transduction model replying entirely on self-attention mechanisms without using sequence-aligned recurrent neural networks (RNNs) or convolution, and it achieved significant improvements for sequence to tasks. The introduced intensive computation storage of these pre-trained representations has impeded their popularity into memory constrained devices. field-programmable gate array (FPGA) is widely used...

The explosion of "big data" applications imposes severe challenges speed and scalability on traditional computer systems. As the performance Von Neumann machines is greatly hindered by increasing gap between CPU memory ("known as wall"), neuromorphic computing systems have gained considerable attention. biology-plausible paradigm carries out emulating charging/discharging process neuron synapse potential. unique spike domain information encoding enables asynchronous event driven computation...

Spiking neural network (SNN) has drawn research interests as it mimics dynamic activities of human brain and the potential to perform real-time cognitive tasks. However, latency, throughput flexibility existing hardware implemented SNNs are limited. The conventional rate coding is inefficient in terms accuracy latency. Oversimplified SNN models adopted by neuromorphic discard characteristics such neuron dynamics filter effects etc., which critical for information processing. Recent...

The recently discovered spatial-temporal information processing capability of bio-inspired Spiking neural networks (SNN) has enabled some interesting models and applications. However designing large-scale high-performance model is yet a challenge due to the lack robust training algorithms. A bio-plausible SNN with property complex dynamic system. Synapses neurons behave as filters capable preserving temporal information. As such neuron dynamics filter effects are ignored in existing...

The trajectory prediction is a critical and challenging problem in the design of an autonomous driving system. Many AI-oriented companies, such as Google Waymo, Uber DiDi, are investigating more accurate vehicle algorithms. However, performance governed by lots entangled factors, stochastic behaviors surrounding vehicles, historical information self-trajectory, relative positions neighbors, etc. In this paper, we propose novel graph-based sharing network (GISNet) that allows between target...

Asynchronous event-driven computation and communication using spikes facilitate the realization of spiking neural networks (SNN) to be massively parallel, extremely energy efficient highly robust on specialized neuromorphic hardware. However, lack a unified learning algorithm limits SNN shallow with low accuracies. Artificial (ANN), however, have backpropagation which can utilize gradient descent train are locally universal function approximators. But is neither biologically plausible nor...

Abstract The rapid growth of technological innovations and human activities has led to significant adverse effects on the environment its quality. To assess ecological quality, this study utilizes a comprehensive indicator known as “load capacity factor,” which incorporates two key components: bio‐capacity footprints. By employing dynamic autoregressive distributed lag (DYARDL) spectral causality approaches, research investigates impact fossil fuel consumption, innovation, renewable energy...

Although widely used in machine learning, backpropagation cannot directly be applied to SNN training and is not feasible on a neuromorphic processor that emulates biological neuron synapses. This work presents spike-based algorithm with plausible local update rules adapts it fit the constraint hardware. The implemented Intel's Loihi chip enabling low power in-hardware supervised online learning of multilayered SNNs for mobile applications. We test this implementation MNIST, Fashion-MNIST,...

Graph Convolutional Networks (GCNs) are pivotal in extracting latent information from graph data across various domains, yet their acceleration on mainstream GPUs is challenged by workload imbalance and memory access irregularity. To address these challenges, we present Accel-GCN, a GPU accelerator architecture for GCNs. The design of Accel-GCN encompasses: (i) lightweight degree sorting stage to group nodes with similar degree; (ii) block-level partition strategy that dynamically adjusts...

Bio-inspired neuromorphic hardware is a research direction to approach brain's computational power and energy efficiency. Spiking neural networks (SNN) encode information as sparsely distributed spike trains employ spike-timingdependent plasticity (STDP) mechanism for learning. Existing implementations of SNN are limited in scale or do not have in-hardware learning capability. In this work, we propose low-cost scalable Network-on-Chip (NoC) based architecture with fully STDP All neurons work...

Abstract Using a database of the trading data in Chinese stock market over January 2005 to June 2012, this paper studies crisis based on perspective behavioural finance. Investor sentiment is B‐W method, and possibility Shanghai was predicted by logit model. The empirical results show that investor sentiment, which more significant than macroeconomic variables, has positive impact after controlling for economic variables. Moreover, our offer an explanation financial anomaly mean reversion....

Abstract The increasing bacterial resistance highlights the necessity of developing superior antibacterial materials. Metal‐doped carbon dots (CDs), a crucial branch carbon‐based nanozymes, exhibit efficient properties using reactive oxygen species (ROS) effect. In this study, iron‐doped (Fe‐CDs) are synthesized with high peroxidase‐like (POD‐like) activity natural hemin as precursor for first time. formation Fe‐CDs remarkably improves water solubility showing excellent biocompatibility....

There is an increasing demand to process streams of temporal data in energy-limited scenarios such as embedded devices, driven by the advancement and expansion Internet Things (IoT) Cyber-Physical Systems (CPS). Spiking neural network has drawn attention it enables low power consumption encoding processing information sparse spike events, which can be exploited for event-driven computation. Recent works also show SNNs' capability spatial information. Such advantages power-limited devices...

Driven by the expanse of Internet Things (IoT) and Cyber-Physical Systems (CPS), there is an increasing demand to process streams temporal data on embedded devices with limited energy power resources. Among all potential solutions, neuromorphic computing spiking neural networks (SNN) that mimic behavior brain, have recently been placed at forefront. Encoding information into sparse distributed spike events enables low-power implementations, complex spatial dynamics synapses neurons enable...

Neuromorphic computing and spiking neural networks (SNN) mimic the behavior of biological systems have drawn interest for their potential to perform cognitive tasks with high energy efficiency. However, some factors such as temporal dynamics spike timings prove critical information processing but are often ignored by existing works, limiting performance applications neuromorphic computing. On one hand, due lack effective SNN training algorithms, it is difficult utilize dynamics. Many...

Continuous trajectory control of fixed-wing unmanned aerial vehicles (UAVs) is complicated when considering hidden dynamics. Due to UAV multi degrees freedom, tracking methodologies based on conventional theory, such as Proportional-Integral-Derivative (PID) has limitations in response time and adjustment robustness, while a model approach that calculates the force torques UAV's current status rigid. We present an actor-critic reinforcement learning framework controls through set desired...

Bio-inspired neuromorphic hardware is an emerging computing architecture, which features highly parallel and distributed elements similar to the functionality of human brain. Recent study shows that can achieve state-of-the-art performance in various cognitive tasks. However, limitations fabrication technology has led fan-in, fan-out, memory capacity, connectivity etc., making chips difficult program. Neural networks have satisfy specific constraints order be mapped hardware, not only...

Deep reinforcement learning (DRL) has been applied for optimal control of autonomous UAV trajectory generation. The energy and payload capacity small UAVs impose constraints on the complexity size neural network. While Model compression potential to optimize trained network model efficient deployment em-bedded platforms, pruning a DRL is more difficult due slow convergence in training before after pruning. In this work, we focus improving speed New reward function action exploration are...

In this work, we develop a novel MOSFET model for circuit simulation purpose. Instead of using traditional physics-driven model, such as BSIM our work uses ANN to the electrical behavior transistor. With unique pre and post-processing procedures, is trained drain current precisely under various applied voltage, device size temperature. The further successfully implemented in SPICE through Verilog-A language. Both n-type p-type MOSFETs show good fitting between models. Eventually, an inverter...

Small Unmanned Aircraft Systems (sUAS) will be an important component of the smart city and intelligent transportation environments near future. The demand for sUAS related applications, such as commercial delivery land surveying, is expected to grow rapidly in next few years. In general, traffic scheduling management functions are needed coordinate launching from different launch sites plan their trajectories avoid conflict while considering several other constraints arrival time, minimum...

The recent discovered spatial-temporal information processing capability of bio-inspired Spiking neural networks (SNN) has enabled some interesting models and applications. However designing large-scale high-performance model is yet a challenge due to the lack robust training algorithms. A bio-plausible SNN with property complex dynamic system. Each synapse neuron behave as filters capable preserving temporal information. As such dynamics filter effects are ignored in existing algorithms,...

In natural language processing (NLP), the "Transformer" architecture was proposed as first transduction model replying entirely on self-attention mechanisms without using sequence-aligned recurrent neural networks (RNNs) or convolution, and it achieved significant improvements for sequence to tasks. The introduced intensive computation storage of these pre-trained representations has impeded their popularity into memory-constrained devices. field-programmable gate array (FPGA) is widely used...