Energy-efficient spikformer has been proposed by integrating the biologically plausible spiking neural network (SNN) and artificial transformer, whereby self-attention (SSA) is used to achieve both higher accuracy lower computational cost. However, it seems that not always necessary, especially in sparse spike-form calculation manners. In this article, we innovatively replace vanilla SSA (using dynamic bases calculating from Query Key) with Fourier transform, wavelet their combinations fixed...

Understanding how erosion processes respond to climatic and tectonic variations is a key issue in geosciences. The Himalaya-Karakoram mountains, with their high elevations, relief, significant water discharge that feeds deep-sea turbidite fans, offer prime setting for deciphering the relative controls on erosion. However, extent which climate change has influenced patterns of this tectonically active region remains debated, partly due limited detailed record its erosional history during...

Machine translation systems have been widely adopted in our daily life, making life easier and more convenient. Unfortunately, erroneous translations may result severe consequences, such as financial losses. This requires to improve the accuracy reliability of machine systems. However, it is challenging test because complexity intractability underlying neural models. To tackle these challenges, we propose a novel metamorphic testing approach by syntactic tree pruning (STP) validate Our key...

Machine translation systems have been widely adopted in our daily life, making life easier and more convenient. Unfortunately, erroneous translations may result severe consequences, such as financial losses. This requires to improve the accuracy reliability of machine systems. However, it is challenging test because complexity intractability underlying neural models. To tackle these challenges, we propose a novel metamorphic testing approach by syntactic tree pruning (STP) validate Our key...

Recently, with the application of deep learning in remote sensing image (RSI) field, classification accuracy RSI has been dramatically improved compared traditional technology. However, even state-of-the-art object recognition convolutional neural networks are fooled by universal adversarial perturbation (UAP). The research on UAP is mostly limited to ordinary images, and RSIs have not studied. To explore basic characteristics UAPs RSIs, this paper proposes a novel method combining an...

Python has gradually become one of the most important programming languages through artificial intelligence's development. PIP, a package management tool for Python, offers one-click installation, allowing developers to utilize other people's code speed up However, any registered member can easily upload packages repository that stores third-party packages. This functionality is used by attackers poison index, i.e., publish enormous malicious pip installing backdoor, gathering information,...

Video object recognition for UAV ground detection is widely used in target search, daily patrol, environmental reconnaissance, and other fields. So, we propose the novel parallel deep learning network with ability of global local joint feature extraction video detection. This paper focuses on solving problems background discrimination required by discovery to realize discovery. Break through key real-time recognition, such as multiscale targets, high complexity, many small dense arrangement,...

Energy-efficient spikformer has been proposed by integrating the biologically plausible spiking neural network (SNN) and artificial Transformer, whereby Spiking Self-Attention (SSA) is used to achieve both higher accuracy lower computational cost. However, it seems that self-attention not always necessary, especially in sparse spike-form calculation manners. In this paper, we innovatively replace vanilla SSA (using dynamic bases calculating from Query Key) with Fourier Transform, Wavelet...

Natural Language Understanding (NLU) aims to make sense of language by enabling computers comprehend text in semantic level, which is a fundamental but challenging task natural processing. Recently, BERT has achieved state-of-the-art performances NLU utilizing pretraining and fine-tuning techniques capture the task-specific information without any structure. However, existing models still cannot context-specific information, leading poor performance when dealing with pervasive ambiguity...

With the rapid development of artificial intelligence technology, dangerous items detection and face industry has gradually become mainstream application if you are using a general-purpose processor for image data processing which is complete, it will produce certain time, there be single function, low computing power, efficiency so on. In this paper, KPU neural network accelerator based on K210 chip designed. The model calculation completed by YOLO network, functions quickly realized....

This project designs an intelligent banking management system.The essence of system is information based on computer technology, communication sensing automatic control technology and video technology.It integrates the entrance to bank building, staff security department into one network, introduces database dangerous goods people.This precise processing center K210 other rough modules, which will be connected with police command center. The network become important guarantee means prevent...

Recently, with the application of deep learning in remote sensing image (RSI) field, classification accuracy RSI has been dramatically improved compared traditional technology. However, even state-of-the-art object recognition convolutional neural networks are fooled by universal adversarial perturbation (UAP). The research on UAP is mostly limited to ordinary images, and RSIs have not studied. To explore basic characteristics UAPs RSIs, this paper proposes a novel method combining an...