We propose Uncertainty Augmented Context Attention network (UACANet) for polyp segmentation which considers an uncertain area of the saliency map. construct a modified version U-Net shape with additional encoder and decoder compute map in each bottom-up stream prediction module propagate to next module. In module, previously predicted is utilized foreground, background we aggregate feature three maps representation. Then relation between representation pixel conduct experiments on five...

Personalized recommendations are the backbone machine learning (ML) algorithm that powers several important application domains (e.g., ads, e-commerce, etc) serviced from cloud datacenters. Sparse embedding layers a crucial building block in designing yet little attention has been paid properly accelerating this ML algorithm. This paper first provides detailed workload characterization on personalized and identifies two significant performance limiters: memory-intensive compute-intensive...

Integrating deep learning techniques, particularly language models (LMs), with knowledge representation techniques like ontologies has raised widespread attention, urging the need of a platform that supports both paradigms. Although packages such as OWL API and Jena offer robust support for basic ontology processing features, they lack capability to transform various types information within into formats suitable downstream learning-based applications. Moreover, widely-used APIs are...

Current image-to-image translations do not control the output domain beyond classes used during training, nor they interpolate between different domains well, leading to implausible results. This limitation largely arises because labels consider semantic distance. To mitigate such problems, we propose a style-aware discriminator that acts as critic well style encoder provide conditions. The learns controllable space using prototype-based self-supervised learning and simultaneously guides...

We propose Spatio-Temporal SlowFast Self-Attention network for action recognition. Conventional Convolutional Neural Networks have the advantage of capturing local area data. However, to understand a human action, it is appropriate consider both and overall context given scene. Therefore, we repurpose self-attention mechanism from GAN (SAGAN) our model retrieving global semantic when making Using mechanism, module that can extract four features in video information: spatial information,...

Moving object detection is very important for video surveillance in modern days. In special case, we can categorize motions into two types - salient and non-salient motion. this paper, first calculate temporal difference image extract moving objects adapt to dynamic environments next, also propose a new algorithm detect motion information complex environment by combining binary block which calculated vector using the newest MPEG-4 EPZS.

With shortening product lifecycles, design is more strongly influencing successful business' competitive advantage. The Kano model has been proposed to define and extract customers' needs for attractive quality creation in development. Because or service differentiation crucial business successes, the extraction of delighter an important issue. This study proposes a methodology extracting factors by using big data outside organisation; such factor may not be megatrend but model. defined as...

Cache side-channel attack extracts secret information by monitoring cache behavior of a victim. Normally, this targets L3 which is shared between spy and Hence, can learn about without perception the To resist against attack, many detection techniques have been proposed in literature. However, those approaches limitation since they do not operate real time. In paper, we propose realtime technique attacks. The performs attacks immediately after observing variation CPU counters. For this,...

Recent operating systems (OSs) have adopted a defense mechanism called kernel page table isolation (KPTI) for protecting the from all attacks that break address space layout randomization (KASLR) using various side-channel analysis techniques. In this paper, we demonstrate KASLR can still be broken, even with latest OSs where KPTI is applied. particular, present novel memory-sharing-based attack breaks on KPTI-enabled Linux virtual machines. The proposed leverages memory deduplication...

Moving object detection is very important for video surveillance in modern days. In special case, we can categorize motions into two types - salient and nonsalient motion. this paper, first calculate temporal difference image extract moving objects adapt to dynamic environments next, also propose a new algorithm detect motion information complex environment by combining binary block which calculated vector using the newest MPEG-4 EPZS

Though science and technology are evolving rapidly in recent years, the traditional education has limits for students to be satisfied their interests needs because they couldn't follow these speeds. STEAM as a integrating science, technology, engineering, arts mathematics strengths of increasing understandings improving integrated thinking problem solving ability leaners. In this study we analyze elementary school curriculum construct physics learning based on develop android application...

Personalized recommendations are the backbone machine learning (ML) algorithm that powers several important application domains (e.g., ads, e-commerce, etc) serviced from cloud datacenters. Sparse embedding layers a crucial building block in designing yet little attention has been paid properly accelerating this ML algorithm. This paper first provides detailed workload characterization on personalized and identifies two significant performance limiters: memory-intensive compute-intensive...

본 논문에서는 약병의 크기와 색상정보 특징을 추출하여 약병영상 분류 기법을 제안한다. 분류에 있어 유사한 모양을 지닌 약병이 다양하게 존재하므로, 한 가지 특징만으로는 약병을 분류하기가 어렵다. 이러한 문제를 해결하기 위해 색상정보의 분류하였다. 제안된 알고리즘의 첫 번째 단계에서는 약병영상에서 Red, Green, Blue의 이진화 문턱치(Binary threshold)를 이용하여 약병 영역의 MBR(Minimum Boundary Rectangle)을 크기로 분류하였고, 두 분류된 가운데 조명의 조도 변화에 강인한 색상(Hue)정보와 RGB 각각의 채널에 대한 컬러 평균 비율 정보를 분류하였으며, 마지막 SURF(Speeded Up Robust Features)알고리즘을 사용하여 데이터베이스에서 특징점을 추출한 후보군 약병영상과 입력 약병영상의 유사도가 가장 높은 약병영상을 검색하여 실험을 통해 방법이 보다 효율적이고 신뢰성 있음을 입증하였다. In this paper,...

Modern computer systems take advantage of Input/Output Memory Management Unit (IOMMU) to protect memory from DMA attacks, or achieve strong isolation in virtualization. Despite its promising benefits, the IOMMU could be a new source security threats. Like MMU, also has Translation Lookaside Buffer (TLB) named IOTLB, an address translation cache that keeps recent translations. Accordingly, IOTLB can target timing side-channel attack, revealing victim's secret. In this paper, we present...

Semantic segmentation networks adopt transfer learning from image classification which occurs a shortage of spatial context information. For this reason, we propose Spatial Context Memoization (SpaM), bypassing branch for by retaining the input dimension and constantly communicating its rich semantic information mutually with backbone network. Multi-scale is crucial dealing diverse sizes shapes target objects in given scene. Conventional multi-scale scheme adopts multiple effective receptive...

Given a set of heterogeneous source datasets with their classifiers, how can we quickly find the most useful dataset for specific target task? We address problem measuring transferability between and datasets, where have different feature spaces distributions. propose Transmeter, fast accurate method to estimate two multivariate datasets. three challenges in datasets: reducing time, minimizing domain gap, extracting meaningful homogeneous representations. To overcome above issues, utilize...

Current image-to-image translations do not control the output domain beyond classes used during training, nor they interpolate between different domains well, leading to implausible results. This limitation largely arises because labels consider semantic distance. To mitigate such problems, we propose a style-aware discriminator that acts as critic well style encoder provide conditions. The learns controllable space using prototype-based self-supervised learning and simultaneously guides...

Salient object detection (SOD) has been in the spotlight recently, yet studied less for high-resolution (HR) images. Unfortunately, HR images and their pixel-level annotations are certainly more labor-intensive time-consuming compared to low-resolution (LR) annotations. Therefore, we propose an image pyramid-based SOD framework, Inverse Saliency Pyramid Reconstruction Network (InSPyReNet), prediction without any of datasets. We design InSPyReNet produce a strict pyramid structure saliency...