Speculative execution is crucial in enhancing modern processor performance but can introduce Spectre-type vulnerabilities that may leak sensitive information.Detecting Spectre gadgets from programs has been a research focus to enhance the analysis and understanding of attacks.However, one problems existing approaches they rely on presence source code (or are impractical terms run-time gadget detection ability).This paper presents Teapot, first scanner works COTS binaries with comparable...

Existing convolutional neural network (CNN)-based methods have limitations in long-term multi-damage recognition for civil infrastructures. Owing to catastrophic forgetting, the accuracy of such networks decreases when structural damage types keep increasing progressively, not mention other issues as an increased number model parameters and data storage. Thus, this study proposes a continual-learning-based (CLDRM) relevant components By integrating Learning without Forgetting (LwF) method...

Recent state-of-the-art Learned Image Compression methods feature spatial context models, achieving great rate-distortion improvements over hyperprior methods. However, the autoregressive model requires serial decoding, limiting run-time performance. The Checkerboard allows parallel decoding at a cost of reduced RD We present series multistage models allowing both fast and better split latent space into square patches decode serially within each patch while different are decoded in parallel....

In this article, local relationship between per capita health care expenditure (HCE hereafter) and GDP is investigated with quantile regressions. Logarithmic HCE of 154 countries in 2001 25, 50 75% regressions are considered. Three main findings obtained from our empirical study. First, conditional distribution on asymmetric. For lower countries, the skewed to right which means less tends be consumed. On contrary, left implies more apt consumed for high countries. Second, variance larger...

Structures suffer from the emergence of cracks, therefore, crack detection is always an issue with much concern in structural health monitoring. Along rapid progress deep learning technology, image semantic segmentation, active research field, offers another solution, which more effective and intelligent, to Through numerous artificial neural networks have been developed address preceding issue, corresponding explorations are never stopped improving quality detection. This paper presents a...

Recently, learned image compression (LIC) has shown a superior ability in the ratio as well quality of reconstructed image. By adopting framework variational autoencoder, LIC [1] can outperform intra prediction latest traditional coding standard VVC. To accelerate speed, most frameworks are operated on GPU with floating-point arithmetic. However, mismatch calculation results various hardware platforms will cause decoding error if encoding and performed different platforms. Therefore,...

The generalized estimating equations (GEE) method is a popular approach for analyzing dependent data of various types.While GEE estimators are robust against the misspecification correlation matrix, their estimation efficiency can be seriously affected by choice working matrix.For spatially correlated data, it difficult to specify true spatial structure due complexity dependence and high dimension matrices.To achieve while allowing flexibility capture complex dependence, we propose new...

Multi-damage is common in reinforced concrete structures and leads to the requirement of large number neural networks, parameters data storage, if convolutional network (CNN) used for damage recognition. In addition, conventional CNN experiences catastrophic forgetting training inefficiency as tasks increases during continual learning, leading accuracy decrease previous learned tasks. To address these problems, this study proposes a continuallearning-based recognition model (CLDRM) which...

In terms of building structural analysis, a large number researchers throw the spotlight on data swapping from Industry Foundation Classes (IFC) models to input files various kinds Finite-Element-Analysis software.These attempts succeeded in accelerating pre-processing before simulation.However, associated methodologies do not break this restriction data-to-data mapping.Up now, an openBIM based mapping methodology describing software-independent form IFC model Finite-Element system emerges...

To improve the efficiency and reduce labour cost of renovation process, this study presents a lightweight Convolutional Neural Network (CNN)-based architecture to extract crack-like features, such as cracks joints. Moreover, Transfer Learning (TF) method was used save training time while offering comparable prediction results. For three different objectives: 1) Detection concrete cracks; 2) natural stone 3) Differentiation between joints in stone; We built dataset with information...

This demo paper gives a real-time learned image codec on FPGA. By using Xilinx VCU128, the proposed system reaches 720P@30fps codec, which is 7.76x faster than prior work.

Recent state-of-the-art Learned Image Compression methods feature spatial context models, achieving great rate-distortion improvements over hyperprior methods. However, the autoregressive model requires serial decoding, limiting runtime performance. The Checkerboard allows parallel decoding at a cost of reduced RD We present series multistage models allowing both fast and better split latent space into square patches decode serially within each patch while different are decoded in parallel....

Entropy coding is essential to data compression, image and video coding, etc.The Range variant of Asymmetric Numeral Systems (rANS) a modern entropy coder, featuring superior speed compression rate.As rANS not designed for parallel execution, the conventional approach partitions input symbol sequence encodes with independent codecs, more bring extra overhead.This found in state-of-the-art implementations such as DietGPU.It unsuitable content-delivery applications, parallelism wasted if...

Learned Image Compression (LIC), which uses neural networks to compress images, has experienced significant growth in recent years. The hyperprior-module-based LIC model achieved higher performance than classical codecs. However, the models are too heavy (in calculation and parameter amounts) apply edge devices. To solve this problem, some former papers focus on structural pruning for models. they either cause noticeable decrement or neglect appropriate threshold each model. These problems...

In various applications with large spatial regions, the relationship between response variable and covariates is expected to exhibit complex patterns. We propose a spatially clustered varying coefficient model, where regression coefficients are allowed vary smoothly within each cluster but change abruptly across boundaries of adjacent clusters, we develop unified approach for simultaneous estimation identification. The approximated by penalized splines, clusters identified through fused...

Semantic segmentation profits from deep learning and has shown its possibilities in handling the graphical data on-site inspection. As a result, visual damage facade images should be detected. Attention mechanism generative adversarial networks are two of most popular strategies to improve quality semantic segmentation. With specific focuses on these strategies, this paper adopts U-net, representative convolutional neural network, as primary network presents comparative study steps. First,...

Learned image compression allows achieving state-of-the-art accuracy and ratios, but their relatively slow runtime performance limits usage. While previous attempts on optimizing learned codecs focused more the neural model entropy coding, we present an alternative method to improving of various models. We introduce multi-threaded pipelining optimized memory enable GPU CPU workloads asynchronous execution, fully taking advantage computational resources. Our architecture alone already...

Learned image compression allows achieving state-of-the-art accuracy and ratios, but their relatively slow runtime performance limits usage. While previous attempts on optimizing learned codecs focused more the neural model entropy coding, we present an alternative method to improving of various models. We introduce multi-threaded pipelining optimized memory enable GPU CPU workloads' asynchronous execution, fully taking advantage computational resources. Our architecture alone already...