In recent years,deep learning methods have been introduced for segmentation and classi-fication of leaf lesions caused by pests and *** the commonly used approaches,convolutional neural networks have provided results ...
详细信息
In recent years,deep learning methods have been introduced for segmentation and classi-fication of leaf lesions caused by pests and *** the commonly used approaches,convolutional neural networks have provided results with high *** purpose of this work is to present an effective and practical system capable of seg-menting and classifying different types of leaf lesions and estimating the severity of stress caused by biotic agents in coffee leaves using convolutional neural *** proposed approach consists of two stages:a semantic segmentation stage with severity calculation and a symptom lesion classification *** stage was tested separately,highlighting the positive and negative points of each *** obtained very good results for the severity estimation,suggesting that the model can estimate severity values very close to the real *** the biotic stress classification,the accuracy rates were greater than 97%.Due to the promising results obtained,an App for Android platform was developed and imple-mented,consisting of semantic segmentation and severity calculation,as well as symptom classification to assist both specialists and farmers to identify and quantify biotic stresses using images of coffee leaves acquired by smartphone.
Approximate computing (AxC) emerged as a design alternative to boost design efficiency by leveraging the inherent error resiliency of many applications. Recent literature shows attackers could exploit some AxC mechani...
Approximate computing (AxC) emerged as a design alternative to boost design efficiency by leveraging the inherent error resiliency of many applications. Recent literature shows attackers could exploit some AxC mechanisms to create new attack surfaces. This work proposes integrity checking and exclusive logic-based methods detect attacks by hardware faults on AxC systems. During ASIC-based synthesis, we investigate the security vulnerabilities of the Approximate Parallel Prefix Adder (AxPPA). We compare the expected output of the designer with the output netlists of ASIC-based synthesis. We propose a verification methodology consisting of a golden model (GM) described in a Simulink model and the DUT represented at the netlist level in a standard cell library after logic synthesis. Experimental results show that the AxPPA has an error range of 1.263 × 10 −5 to 0.0098 concerning the designer’s expected output under ASIC-based synthesis attacks.
Propranolol hydrochloride can be considered a persistent and bioaccumulative pharmaceutical in the *** drug and its by-products are potentially toxic and have adverse effects,since these compounds have been associated...
详细信息
Propranolol hydrochloride can be considered a persistent and bioaccumulative pharmaceutical in the *** drug and its by-products are potentially toxic and have adverse effects,since these compounds have been associated with endocrine-disrupting effects,reproductive deficiencies,embryo abnormalities and pericardial ***_(2)–La 0.05%–carboxymethyl-β-cyclodextrin(CMCD)nanoparticles were successfully prepared by a simple two-step method,which consists of sonification and *** characterization analyses reveal that lanthanum is dispersed on the semiconductor surface,probably forming Ti–O–La bonds,which can induce oxygen vacancies and surface defects that effectively restrain the recombination of photogenerated electron/holes *** efficiency of TiO_(2)–La 0.05%–CMCD samples in degradation of propranolol under UV-light irradiation is higher than that of pristine TiO_(2) within 20 min reaction,probably due to complex formation between theβ-blocker and the oligosaccharide,which allows us to propose a photocatalytic mechanism based on the formation of intermediates and competition of these compounds to the radicals and CMCD cavities.
Addition units are present across many computational kernels inherent in various error-tolerant applications, including machine learning, signal, image, and video processing. Notably, adder compressors are the target ...
Addition units are present across many computational kernels inherent in various error-tolerant applications, including machine learning, signal, image, and video processing. Notably, adder compressors are the target when high speed and low power are the main design concerns. This work introduces energy-efficient structures of 3-2 approximate adder compressors (Ax3-2) and three distinct versions of 4-2 approximate adder compressors (Ax4-2). We compared our proposed Ax3-2 and Ax4-2 compressors with state-of-the-art energy-efficient approximate adder compressors (AxACs). Both Ax3-2 and Ax4-2 models passed through rigorous tests as standalone units. Furthermore, we integrated the Ax4-2 models into an essential application kernel appearing in video coding, i.e., the Sum Squared Differences (SSD) video accelerator. We introduce Ax3-2 and Ax4-2 that yield a new Pareto front concerning both energy-quality and area-quality trade-offs, thereby demonstrating a marked improvement over the prevailing state-of-the-art energy-efficient AxACs.
This work proposes an ultra-low-energy ECG data compression with VLSI DHWT-based (discrete Haar wavelet transform) architecture to enable storage and transmission in resource-constrained environments. We present origi...
This work proposes an ultra-low-energy ECG data compression with VLSI DHWT-based (discrete Haar wavelet transform) architecture to enable storage and transmission in resource-constrained environments. We present original, pruned, and approximate DHWT (ODHWT, PDHWT, and AxDHWT, respectively) hardware architectures for ECG data compression at ultra-high energy efficiency. Our best proposal employing the AxDHWT hardware architecture requires just five additions only. Using a PDHWT technique to improve energy efficiency observes the evolution of the signal-to-noise ratio and the ultimate impact on the ECG data compression application. The DHWT-based configurations architecture proposal achieves a minimum compression ratio of 0.125 (i.e., 1/8) and a PRD (percent root difference) <1.34. We implemented the VLSI architectures in 65nm CMOS technology in maximum and fixed frequencies and two different supply voltages. The AxDHWT occupies a die area of 0.64mm 2 . The measured total power is $0.534\mu {\mathrm {W}}$, the higher energy-savings among the ECG data compression architectures.
This work proposes an accuracy-driven evaluation model of an approximate adder (AxA) called an approximate parallel prefix adder (AxPPA). It explores the significance of approximate computing (AxC) in optimizing perfo...
This work proposes an accuracy-driven evaluation model of an approximate adder (AxA) called an approximate parallel prefix adder (AxPPA). It explores the significance of approximate computing (AxC) in optimizing performance and energy efficiency in computer systems by accepting minor accuracy trade-offs. We examine the AxPPA through a meticulous error analysis model and an automated analysis framework that comprehensively evaluates area and delay based on the fundamental equations of parallel prefix adders (PPAs). Additionally, it emphasizes the importance of lower-part approximation in adders, effectively addressing the elongation of carry chains and enhancing overall efficiency. The choice of approximation method directly impacts computational accuracy, resulting in variations in the number of approximated bits and carry chain length for a given tolerance level. Conducting a thorough comparison of accuracy trade-offs requires a unique design and simulation for each unit and approximation level, which can be time-consuming, particularly in larger circuits. Therefore, the proposed approach facilitates the exploration of larger circuit design space encompassing multiple AxAs, significantly minimizing time investment. In our experiments, the accuracy-driven evaluation of the lower part approximation method in AxA took 71.268 milliseconds for 1 million random inputs, with a K -bits approximation ranging from 1 to 16. The results indicate that AxPPA exhibits a minor error magnitude among the explored literature AxAs, being approximately 50.97% more accurate on average.
The 5G Quality-of-Service (QoS) objectives con-tributed to the Heterogeneous Cellular Network (HCN) evolution, dictating that applications can rely on low-latency and high-bandwidth networks. However, concurrent reque...
详细信息
ISBN:
(数字)9781665497923
ISBN:
(纸本)9781665497930
The 5G Quality-of-Service (QoS) objectives con-tributed to the Heterogeneous Cellular Network (HCN) evolution, dictating that applications can rely on low-latency and high-bandwidth networks. However, concurrent requests of large amount of multimedia data generate a burden on the backhaul and fronthaul networks due to redundant retransmissions and pose challenges for achieving the QoS objectives. Although mobile network operators can place content closer to the HCN edge to improve the overall QoS indicators, there are still challenges to design a cache policy aware of limited storage capacity, different content popularity, device mobility, and network congestion. This work innovates by introducing a cooperative policy to join caches placement and routing users' requests atop an HCN. By combining networking and cache QoS requirements, the policy balances the fronthaul network load and dynamically maps the caches to HCN resources. We formulated the cache policy through linear programming and in-depth evaluated its performance using extensive simulation scenarios. The results indicate that the proposed network-aware policy decreases the network latency, even when subject to changes in content popularity distribution and total HCN storage capacity.
Equipment monitoring for failure prediction is receiving attention from different sectors of society, such as industry, healthcare, and defense. In the defense domain, assets like military vehicles generate data that ...
详细信息
Out-of-distribution (OOD) detectors can act as safety monitors in embedded cyber-physical systems by identifying samples outside a machine learning model’s training distribution to prevent potentially unsafe actions....
详细信息
ISBN:
(数字)9798350387957
ISBN:
(纸本)9798350387964
Out-of-distribution (OOD) detectors can act as safety monitors in embedded cyber-physical systems by identifying samples outside a machine learning model’s training distribution to prevent potentially unsafe actions. However, OOD detectors are often implemented using deep neural networks, which makes it difficult to meet real-time deadlines on embedded systems with memory and power constraints. We consider the class of variational autoencoder (VAE) based OOD detectors where OOD detection is performed in latent space, and apply quantization, pruning, and knowledge distillation. These techniques have been explored for other deep models, but no work has considered their combined effect on latent space OOD detection. While these techniques increase the VAE’s test loss, this does not correspond to a proportional decrease in OOD detection performance and we leverage this to develop lean OOD detectors capable of real-time inference on embedded CPUs and GPUs. We propose a design methodology that combines all three compression techniques and yields a significant decrease in memory and execution time while maintaining AUROC for a given OOD detector. We demonstrate this methodology with two existing OOD detectors on a Jetson Nano and reduce GPU and CPU inference time by 20% and 28% respectively while keeping AUROC within 5% of the baseline.
H.266/NVC is the current state-of-the-art video coding format, launched in 2020. Despite its high capability to compress video, including ultra-high definition content, royalty costs may prevent its wide adoption. The...
H.266/NVC is the current state-of-the-art video coding format, launched in 2020. Despite its high capability to compress video, including ultra-high definition content, royalty costs may prevent its wide adoption. The AOMedia Video 1 (AV1) format is an alternative with competitive coding efficiency, besides being a royalty-free format. However, migrating legacy content from one format to another is a costly task, which requires long processing times. This work presents a solution for accelerating the H.266/NVC-to-AV1 trans coding based on machine learning. Twelve decision tree models trained with data gathered during the H.266NVC decoding and the AV1 encoding processes are proposed and implemented in the libaom reference software, leading to a complexity reduction of 12.60 % at the cost of coding efficiency losses of 1.81 % on average. To the best of the authors' knowledge, this is the first H.266NVC-to-AV1 trans coding acceleration solution published in the literature.
暂无评论