On-device training for deep neural networks(DNN) has become a trend due to various user preferences and scenarios. The DNN training process consists of three phases, feedforward(FF), backpropagation(BP), and weight gr...
详细信息
On-device training for deep neural networks(DNN) has become a trend due to various user preferences and scenarios. The DNN training process consists of three phases, feedforward(FF), backpropagation(BP), and weight gradient(WG) update. WG takes about one-third of the computation in the whole training process. Current training accelerators usually ignore the special computation property of WG and process it in a way similar to FF/BP. Besides, the extensive data sparsity existing in WG, which brings opportunities to save computation, is not well explored. Nevertheless, exploiting the optimization opportunities would meet three underutilization problems, which are caused by(1) the mismatch between WG data dimensions and hardware parallelism,(2) the full sparsity, i.e., the sparsity of feature map(Fmap),error map(Emap), and gradient, and(3) the workload imbalance resulting from irregular sparsity. In this paper, we propose a specific architecture for sparse weight gradient(SWG) computation. The architecture is designed based on hierarchical unrolling and sparsity-aware(HUSA) dataflow to exploit the optimization opportunities of the special computation property and full data sparsity. In HUSA dataflow, the data dimensions are unrolled hierarchically on the hardware architecture. A valid-data trace(VDT) mechanism is embedded in the dataflow to avoid the underutilization caused by the two-sided input sparsity. The gradient is unrolled in PE to alleviate the underutilization induced by output sparsity while maintaining the data reuse opportunities. Besides, we design an intra-and inter-column balancer(IIBLC) to dynamically tackle the workload imbalance problem resulting from the irregular sparsity. Experimental results show that with HUSA dataflow exploiting the full sparsity, SWG achieves a speedup of 12.23× over state-of-the-art gradient computation architecture, Train Ware. SWG helps to improve the energy efficiency of the state-of-the-art training accelerator LNPU from
Deep neural networks are commonly used in computer vision tasks,but they are vulnerable to adversarial samples,resulting in poor recognition *** traditional algorithms that craft adversarial samples have been effectiv...
详细信息
Deep neural networks are commonly used in computer vision tasks,but they are vulnerable to adversarial samples,resulting in poor recognition *** traditional algorithms that craft adversarial samples have been effective in attacking classification models,the attacking performance degrades when facing object detection models with more complex *** address this issue better,in this paper we first analyze the mechanism of multi-scale feature extraction of object detection models,and then by constructing the object feature-wise attention module and the perturbation extraction module,a novel adversarial sample generation algorithm for attacking detection models is ***,in the first module,based on the multi-scale feature map,we reduce the range of perturbation and improve the stealthiness of adversarial samples by computing the noise distribution in the object *** in the second module,we feed the noise distribution into the generative adversarial networks to generate adversarial perturbation with strong attack *** doing so,the proposed approach possesses the ability to better confuse the judgment of detection *** carried out on the DroneVehicle dataset show that our method is computationally efficient and works well in attacking detection models measured by qualitative analysis and quantitative analysis.
We study adaptive control for a family of nonlinear systems, involving unknown and uncertain parameters. The proposed control law estimates the system parameters adaptively and stabilizes the closedloop system asympto...
详细信息
We study adaptive control for a family of nonlinear systems, involving unknown and uncertain parameters. The proposed control law estimates the system parameters adaptively and stabilizes the closedloop system asymptotically for the initial state over any given bounded set of the state-space. Moreover,reconstruction filters are designed to obtain error residue signals and to enable the use of the least-squares algorithm for estimating the parameters, in order to achieve the convergence based on the persistent excitation condition and asymptotic linearization. The proposed methods are applicable to full actuation and under actuation control systems. Simulation studies are carried out for a pendulum system and for a third-order vehicle model, as well as control of vehicle platoons, validating the theoretical results presented in this paper.
Federated recommender systems(FedRecs) have garnered increasing attention recently, thanks to their privacypreserving benefits. However, the decentralized and open characteristics of current FedRecs present at least t...
详细信息
Federated recommender systems(FedRecs) have garnered increasing attention recently, thanks to their privacypreserving benefits. However, the decentralized and open characteristics of current FedRecs present at least two ***, the performance of FedRecs is compromised due to highly sparse on-device data for each client. Second, the system's robustness is undermined by the vulnerability to model poisoning attacks launched by malicious users. In this paper, we introduce a novel contrastive learning framework designed to fully leverage the client's sparse data through embedding augmentation, referred to as CL4FedRec. Unlike previous contrastive learning approaches in FedRecs that necessitate clients to share their private parameters, our CL4FedRec aligns with the basic FedRec learning protocol, ensuring compatibility with most existing FedRec implementations. We then evaluate the robustness of FedRecs equipped with CL4FedRec by subjecting it to several state-of-the-art model poisoning attacks. Surprisingly, our observations reveal that contrastive learning tends to exacerbate the vulnerability of FedRecs to these attacks. This is attributed to the enhanced embedding uniformity, making the polluted target item embedding easily proximate to popular items. Based on this insight, we propose an enhanced and robust version of CL4FedRec(rCL4FedRec) by introducing a regularizer to maintain the distance among item embeddings with different popularity levels. Extensive experiments conducted on four commonly used recommendation datasets demonstrate that rCL4FedRec significantly enhances both the model's performance and the robustness of FedRecs.
Compared with traditional environments,the cloud environment exposes online services to additional vulnerabilities and threats of cyber attacks,and the cyber security of cloud platforms is becoming increasingly promin...
详细信息
Compared with traditional environments,the cloud environment exposes online services to additional vulnerabilities and threats of cyber attacks,and the cyber security of cloud platforms is becoming increasingly prominent.A piece of code,known as a Webshell,is usually uploaded to the target servers to achieve multiple *** Webshell attacks has become a hot spot in current ***,the traditional Webshell detectors are not built for the cloud,making it highly difficult to play a defensive role in the cloud ***,a Webshell detection system based on deep learning that is successfully applied in various scenarios,is proposed in this *** system contains two important components:gray-box and neural network *** gray-box analyzer defines a series of rules and algorithms for extracting static and dynamic behaviors from the code to make the decision *** neural network analyzer transforms suspicious code into Operation Code(OPCODE)sequences,turning the detection task into a classification *** experiment results show that SmartEagleEye achieves an encouraging high detection rate and an acceptable false-positive rate,which indicate its capability to provide good protection for the cloud environment.
Underwater target detection is an important method for detecting marine organisms. However, due to the image occlusion of underwater targets, blurred water quality, poor lighting conditions, small targets, and complex...
详细信息
Dear Editor,This letter presents a new transfer learning framework for the deep multi-agent reinforcement learning(DMARL) to reduce the convergence difficulty and training time when applying DMARL to a new scenario [1...
详细信息
Dear Editor,This letter presents a new transfer learning framework for the deep multi-agent reinforcement learning(DMARL) to reduce the convergence difficulty and training time when applying DMARL to a new scenario [1], [2].
Understanding and predicting air quality is pivotal for public health and environmental management, especially in urban areas like Delhi. This study utilizes a comprehensive dataset from the Central Pollution Control ...
详细信息
Although Convolutional Neural Networks(CNNs)have achieved remarkable success in image classification,most CNNs use image datasets in the Red-Green-Blue(RGB)color space(one of the most commonly used color spaces).The e...
详细信息
Although Convolutional Neural Networks(CNNs)have achieved remarkable success in image classification,most CNNs use image datasets in the Red-Green-Blue(RGB)color space(one of the most commonly used color spaces).The existing literature regarding the influence of color space use on the performance of CNNs is *** paper explores the impact of different color spaces on image classification using *** compare the performance of five CNN models with different convolution operations and numbers of layers on four image datasets,each converted to nine color *** find that color space selection can significantly affect classification accuracy,and that some classes are more sensitive to color space changes than *** color spaces may have different expression abilities for different image features,such as brightness,saturation,hue,*** leverage the complementary information from different color spaces,we propose a pseudo-Siamese network that fuses two color spaces without modifying the network *** experiments show that our proposed model can outperform the single-color-space models on most *** also find that our method is simple,flexible,and compatible with any CNN and image dataset.
The Internet of Everything(IoE)based cloud computing is one of the most prominent areas in the digital big data *** approach allows efficient infrastructure to store and access big real-time data and smart IoE service...
详细信息
The Internet of Everything(IoE)based cloud computing is one of the most prominent areas in the digital big data *** approach allows efficient infrastructure to store and access big real-time data and smart IoE services from the *** IoE-based cloud computing services are located at remote locations without the control of the data *** data owners mostly depend on the untrusted Cloud Service Provider(CSP)and do not know the implemented security *** lack of knowledge about security capabilities and control over data raises several security *** Acid(DNA)computing is a biological concept that can improve the security of IoE big *** IoE big data security scheme consists of the Station-to-Station Key Agreement Protocol(StS KAP)and Feistel cipher *** paper proposed a DNA-based cryptographic scheme and access control model(DNACDS)to solve IoE big data security and access *** experimental results illustrated that DNACDS performs better than other DNA-based security *** theoretical security analysis of the DNACDS shows better resistance capabilities.
暂无评论