Invited Paper

  • 홈ˆ
  • 행사안내
  • Invited Paper

[Invited Paper] 

한국정보과학회가 선정한 SW분야 최우수학술대회에서 발표된 논문의 저자를 초청하여, 아래와 같이 해당 분야 Oral세션에서 핵심연구내용 소개와 질의응답을 통해 학생들의 연구의욕을 고취하는 프로그램을 마련하였습니다. 관심 있는 분들의 많은 참여 바랍니다.

 

no

분과

학술대회

논문제목

발표자

지도
교수

발표순서

1

고성능
컴퓨팅

CCGrid 2019

DCA-IO: A Dynamic I/O Control Scheme for Parallel and Distributed File System

김성곤(서울대)

엄현상

O2.4-01

2

데이터
베이스

SIGMOD 2019

DistME: A Fast and Elastic Distributed Matrix Computation Engine using GPUs

한동형(DGIST)

김민수

O1.5-01

3

ICML 2019
(oral paper)

SELFIE: Refurbishing Unclean Samples for Robust Deep Learning

송환준(KAIST)

이재길

O3.7-01

4

WWW 2019
(full research paper)

From Small-scale to Large-scale Text Classification

김강민(고려대)

이상근

O3.7-02

5

WWW 2019

RealGraph: A Graph Engine Leveraging the Power-Law Distribution of Real-World Graphs

장명환(한양대)

김상욱

O5.6-01

6

IEEE ICDE 2019

No, That’s Not My Feedback: TV Show Recommendation Using Watchable Interval

이연창(한양대)

김상욱

O3.7-03

7

소프트웨어
공학

ICSE 2018

Automatically generating search heuristics for concolic testing

차수영(고려대)

오학주

O1.1-01

8

FSE 2018

MemFix: Static Analysis-Based Repair of Memory Deallocation Errors for C

이준희(고려대)

오학주

O1.1-02

9

ASE 2018

Template-Guided Concolic Testing via Online Learning

차수영(고려대)

오학주

O1.1-03

10

ICSE 2018

Precise concolic unit testing of C programs using extended units and symbolic alarm filtering

김윤호(KAIST)

김문주

O2.6-01

11

인공지능

ICML 2019

EMI: Exploration with Mutual Information

김형석(서울대)

송현오

O1.6-01

12

ICML 2019

Self-Attention Graph Pooling

이준현(고려대)

강재우

O2.8-01

13

ICML 2019

Training CNNs with Selective Allocation of Channels

정종헌(KAIST)

신진우

O2.8-02

14

CVPR 2019

Deep Metric Learning Beyond Binary Supervision

김성연(포항공대)

곽수하

O3.8-01

15

CVPR 2019

FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stochastic Inference

이정범(서울대)

윤성로

O3.8-02

16

컴퓨터
시스템

ACM SIGMETRICS 2019

TeksDB: Weaving Data Structures for a High-Performance Key-Value Store

한유일(숭실대)

이은지

O1.7-01

17

HPCA 2019

Amoeba: An Autonomous Backup and Recovery SSD for Ransomware Attack Defense

민동현(서강대)

김영재

O2.2-01

18

FAST 2019

Design Tradeoffs for SSD Reliability

김석준(서울대)

민상렬

O2.2-02

19

MobiSys 2019

Graphics-aware Power Governing for Mobile Devices

최용훈(연세대)

차호정

O3.2-01

20

DAC 2019

Fast and Efficient Information Transmission with Burst Spikes in Deep Spiking Neural Networks

박성식(서울대)

윤성로

O3.2-02

21

HPCA 2019

CIDR: A Cost-Effective In-line Data Reduction System for Terabit-per-Second Scale SSD Arrays

Mohammadamin Ajdari(포항공대)

김장우

O5.3-01

22

MobiSys 2019

LpGL: Low-power Graphics Library for Mobile AR Headsets

최재원(아주대)

고정길

O5.3-02

23

컴퓨터
이론

SIGMOD 2019

Efficient Subgraph Matching: Harmonizing Dynamic Programming, Adaptive Matching Order, and Failing Set Together

구건모(서울대)

박근수

O2.5-01

 

[고성능컴퓨팅]

논문제목 : DCA-IO: A Dynamic I/O Control Scheme for Parallel and Distributed File System

저자 : Sunggon Kim, Alex Sim, Kesheng Wu, Suren Byna, Teng Wang, Yongseok Son, Hyeonsang Eom(Seoul National University)

학술대회 : CCGrid 2019

발표자 : 김성곤(서울대)

지도교수 : 엄현상

논문초록 : In high-performance computing, storage is a shared resource and used by all users with many different application requirements and knowledge of storage. Consequently, the optimal storage configuration varies according to the I/O behavior of each application. While system logs are helpful resources in understanding the storage behavior, it is non-trivial for each user to analyze the logs and adjust complex configurations. Even for experienced users, it is difficult to understand the full stack of I/O systems and find the optimal configuration for the specific application. In this work, we analyzed the I/O activities of CORI which is an HPC system in National Energy Research Scientific Computing Center(NERSC). The result of our analysis shows that most users do not adjust storage configurations and use the default settings. Also, it shows that only a few applications are executed repeatedly in the HPC environment. Based on this result, we have developed DCA-IO, a dynamic distributed file system configuration adjustment algorithm, which utilizes system log information and widely adapted rules to adjust storage configurations automatically without any user intervention. DCA-IO utilizes existing system logs and does not require any modifications in code or an additional library. To demonstrate the effectiveness of DCA-IO, we have performed experiments using I/O kernels of the real applications in both isolated small-sized Lustre environment and CORI. Our experimental result shows that the use of our scheme can lead to improvements in the performance of HPC applications by up to 75% in an isolated environment and 50% in a real HPC environment without user intervention.

[데이터베이스]

논문제목 : DistME: A Fast and Elastic Distributed Matrix Computation Engine using GPUs

저자 : Donghyoung Han, Yoon-Min Nam, Jihye Lee(DGIST), Kyongseok Park, Hyunwoo Kim(KISTI), Min-Soo Kim(DGIST)

학술대회 : SIGMOD 2019

발표자 : 한동형(DGIST)

지도교수 : 김민수

논문초록 : Matrix computation, in particular, matrix multiplication is time-consuming, but essentially and widely used in a large number of applications in science and industry. The existing distributed matrix multiplication methods only focus on either low communication cost(i.e., high performance) with the risk of out of memory or large-scale processing with high communication overhead. We propose a distributed elastic matrix multiplication method called CuboidMM that achieves both high performance and large-scale processing. We also propose a GPU acceleration method that can be combined with CuboidMM. CuboidMM partitions matrices into cuboids for optimizing the network communication cost with considering memory usage per task, and the GPU acceleration method partitions a cuboid into subcuboids for optimizing the PCI-E communication cost with considering GPU memory usage. We implement a fast and elastic matrix computation engine called DistME by integrating CuboidMM with GPU acceleration on top of Apache Spark. Through extensive experiments, we have demonstrated that CuboidMM and DistME significantly outperform the state-of-the-art methods and systems, respectively, in terms of both performance and data size.

논문제목 : SELFIE: Refurbishing Unclean Samples for Robust Deep Learning

저자 : Hwanjun Song, Minseok Kim, Jae-Gil Lee(KAIST)

학술대회 : ICML 2019(oral paper)

발표자 : 송환준(KAIST)

지도교수 : 이재길

논문초록 : Owing to the extremely high expressive power of deep neural networks, their side effect is to totally memorize training data even when the labels are extremely noisy. To overcome overfitting on the noisy labels, we propose a novel robust training method called SELFIE. Our key idea is to selectively refurbish and exploit unclean samples that can be corrected with high precision, thereby gradually increasing the number of available training samples. Taking advantage of this design, SELFIE effectively prevents the risk of noise accumulation from the false correction and fully exploits the training data. To validate the superiority of SELFIE, we conducted extensive experimentation using four real-world or synthetic data sets. The result showed that SELFIE remarkably improved absolute test error compared with two state-of-the-art methods.

논문제목 : From Small-scale to Large-scale Text Classification

저자 : Kang-Min Kim, Yeachan Kim, Jungho Lee, Ji-Min Lee, Sangkeun Lee(Korea University)

학술대회 : WWW 2019(full research paper)

발표자 : 김강민(고려대)

지도교수 : 이상근

논문초록 : Neural network models have achieved impressive results in the field of text classification. However, existing approaches often suffer from insufficient training data in a large-scale text classification involving a large number of categories(e.g., several thousands of categories). Several neural network models have utilized multi-task learning to overcome the limited amount of training data. However, these approaches are also limited to small-scale text classification. In this paper, we propose a novel neural network-based multi-task learning framework for large-scale text classification. To this end, we first treat the different scales of text classification(i.e., large and small numbers of categories) as multiple, related tasks. Then, we train the proposed neural network, which learns small- and large-scale text classification tasks simultaneously. In particular, we further enhance this multi-task learning architecture by using a gate mechanism, which controls the flow of features between the small- and large-scale text classification tasks. Experimental results clearly show that our proposed model improves the performance of the large-scale text classification task with the help of the small-scale text classification task. The proposed scheme exhibits significant improvements of as much as 14% and 5% in terms of micro-averaging and macro-averaging F1-score, respectively, over state-of-the-art techniques.

논문제목 : RealGraph: A Graph Engine Leveraging the Power-Law Distribution of Real-World Graphs

저자 : Yong-Yeon Jo, Myung-Hwan Jang, Sang-Wook Kim, Sunju Park(Hanyang University)

학술대회 : WWW 2019

발표자 : 장명환(한양대)

지도교수 : 김상욱

논문초록 : As the size of real-world graphs has drastically increased in recent years, a wide variety of graph engines have been developed to deal with such big graphs efficiently. However, the majority of graph engines have been designed without considering the power-law degree distribution of real-world graphs seriously. Two problems have been observed when existing graph engines process real-world graphs: inefficient scanning of the sparse indicator and the delay in iteration progress due to uneven workload distribution. In this paper, we propose RealGraph, a single-machine based graph engine equipped with the hierarchical indicator and the block-based workload allocation. Experimental results on real-world datasets show that RealGraph significantly outperforms existing graph engines in terms of both speed and scalability.

논문제목 : No, That’s Not My Feedback: TV Show Recommendation Using Watchable Interval

저자 : Kyung-Jae Cho, Yeon-Chang Lee, Kyungsik Han, Jaeho Choi, Sang-Wook Kim(Hanyang University)

학술대회 : IEEE ICDE 2019

발표자 : 이연창(한양대)

지도교수 : 김상욱

논문초록 : As the number of TV channels increases, it is becoming important to recommend TV shows that users prefer to watch. To this end, we investigate the inherent characteristics of implicit feedback given in the TV show domain, and identify the challenges for building an effective TV show recommendation. Based on the unique characteristics, we define a user’s watchable interval, the most important and novel concept in understanding users’ true preferences. In order to reflect this new concept into the TV show recommendation, we propose a novel framework based on collaborative filtering. Our framework is composed of (1) preference estimation based on a user’s watchable interval, (2) preference prediction based on confidence exploiting watchable episodes, and (3) top-N recommendation considering TV show’s staying and remaining times. Using a real-world TV show dataset, we demonstrate that our framework effectively solves the challenges and significantly outperforms other existing state-of-the-art methods.

[소프트웨어공학]

논문제목 : Automatically generating search heuristics for concolic testing

저자 : Sooyoung Cha, Seongjoon Hong, Junhee Lee, Hakjoo Oh(Korea University)

학술대회 : ICSE 2018

발표자 : 차수영(고려대)

지도교수 : 오학주

논문초록 : We present a technique to automatically generate search heuristics for concolic testing. A key challenge in concolic testing is how to effectively explore the program’s execution paths to achieve high code coverage in a limited time budget. Concolic testing employs a search heuristic to address this challenge, which favors exploring particular types of paths that are most likely to maximize the final coverage. However, manually designing a good search heuristic is nontrivial and typically ends up with suboptimal and unstable outcomes. The goal of this paper is to overcome this shortcoming of concolic testing by automatically generating search heuristics. We define a class of search heuristics, namely a parameterized heuristic, and present an algorithm that efficiently finds an optimal heuristic for each subject program. Experimental results with open-source C programs show that our technique successfully generates search heuristics that significantly outperform existing manually-crafted heuristics in terms of branch coverage and bug-finding.

논문제목 : MemFix: Static Analysis-Based Repair of Memory Deallocation Errors for C

저자 : Junhee Lee, Seongjoon Hong, Hakjoo Oh(Korea University)

학술대회 : FSE 2018

발표자 : 이준희(고려대)

지도교수 : 오학주

논문초록 : We present MemFix, an automated technique for fixing memory deallocation errors in C programs. MemFix aims to fix memory-leak, double-free, and use-after-free errors, which occur when developers fail to properly deallocate memory objects. MemFix attempts to fix these errors by finding a set of free-statements that correctly deallocate all allocated objects without causing double-frees and use-after-frees. The key insight behind MemFix is that finding such a set of deallocation statements corresponds to solving an exact cover problem derived from a variant of typestate static analysis. We formally present the technique and experimentally show that MemFix is able to fix real errors found in open-source programs. Because MemFix is based on a sound static analysis, the generated patches guarantee to fix the original error without introducing new errors.

논문제목 : Template-Guided Concolic Testing via Online Learning

저자 : Sooyoung Cha, Seonho Lee, Junhee Lee, Hakjoo Oh(Korea University)

학술대회 : ASE 2018

발표자 : 차수영(고려대)

지도교수 : 오학주

논문초록 : We present template-guided concolic testing, a new technique for effectively reducing the search space in concolic testing. Addressing the path-explosion problem has been a significant challenge in concolic testing. Diverse search heuristics have been proposed to mitigate this problem but using search heuristics alone is not sufficient to substantially improve code coverage for real-world programs. The goal of this paper is to complement existing techniques and achieve higher coverage by exploiting templates in concolic testing. In our approach, a template is a partially symbolized input vector whose job is to reduce the search space. However, choosing a right set of templates is nontrivial and significantly affects the final performance of our approach. We present an algorithm that automatically learns useful templates online, based on data collected from previous runs of concolic testing. The experimental results with open-source programs show that our technique achieves greater branch coverage and finds bugs more effectively than conventional concolic testing.

논문제목 : Precise concolic unit testing of C programs using extended units and symbolic alarm filtering

저자 : Yunho Kim(KAIST), Yunja Choi(Kyungpook National University), Moonzoo Kim(KAIST)

학술대회 : ICSE 2018

발표자 : 김윤호(KAIST)

지도교수 : 김문주

논문초록 : Automated unit testing reduces manual effort to write unit test drivers/stubs and generate unit test inputs. However, automatically generated unit test drivers/stubs raise false alarms because they often over-approximate real contexts of a target function f and allow infeasible executions of f. To solve this problem, we have developed a concolic unit testing technique CONBRIO. To provide realistic context to f. it constructs an extended unit of f that consists of f and closely relevant functions to f. Also, CONBRIO filters out a false alarm by checking feasibility of a corresponding symbolic execution path with regard to f ’s symbolic calling contexts obtained by combining symbolic execution paths of f ’s closely related predecessor functions. In the experiments on the crash bugs of 15 real-world C programs, CONBRIO shows both high bug detection ability(i.e. 91.0% of the target bugs detected) and high precision(i.e. a true to false alarm ratio is 1:4.5). Also, CONBRIO detects 14 new bugs in 9 target C programs studied in papers on crash bug detection techniques.

[인공지능]

논문제목 : EMI: Exploration with Mutual Information

저자 : Hyoungseok Kim, Jaekyeom Kim, Yeonwoo Jeong(Seoul National University), Sergey Levine(UC Berkeley) Hyun Oh Song(Seoul National University)

학술대회 : ICML 2019

발표자 : 김형석(서울대)

지도교수 : 송현오

논문초록 : Reinforcement learning algorithms struggle when the reward signal is very sparse. In these cases, naive random exploration methods essentially rely on a random walk to stumble onto a rewarding state. Recent works utilize intrinsic motivation to guide the exploration via generative models, predictive forward models, or discriminative modeling of novelty. We propose EMI, which is an exploration method that constructs embedding representation of states and actions that does not rely on generative decoding of the full observation but extracts predictive signals that can be used to guide exploration based on forward prediction in the representation space. Our experiments show competitive results on challenging locomotion tasks with continuous control and on image-based exploration tasks with discrete actions on Atari. The source code is available at https://github.com/snu-mllab/EMI.

논문제목 : Self-Attention Graph Pooling

저자 : Junhyun Lee, Inyeop Lee, Jaewoo Kang(Korea University)

학술대회 : ICML 2019

발표자 : 이준현(고려대)

지도교수 : 강재우

논문초록 : Advanced methods of applying deep learning to structured data such as graphs have been proposed in recent years. In particular, studies have focused on generalizing convolutional neural networks to graph data, which includes redefining the convolution and the downsampling(pooling) operations for graphs. The method of generalizing the convolution operation to graphs has been proven to improve performance and is widely used. However, the method of applying downsampling to graphs is still difficult to perform and has room for improvement. In this paper, we propose a graph pooling method based on self-attention. Self-attention using graph convolution allows our pooling method to consider both node features and graph topology. To ensure a fair comparison, the same training procedures and model architectures were used for the existing pooling methods and our method. The experimental results demonstrate that our method achieves superior graph classification performance on the benchmark datasets using a reasonable number of parameters.

논문제목 : Training CNNs with Selective Allocation of Channels

저자 : Jongheon Jeong, Jinwoo Shin(KAIST)

학술대회 : ICML 2019

발표자 : 정종헌(KAIST)

지도교수 : 신진우

논문초록 : Recent progress in deep convolutional neural networks(CNNs) have enabled a simple paradigm of architecture design: larger models typically achieve better accuracy. Due to this, in modern CNN architectures, it becomes more important to design models that generalize well under certain resource constraints, e.g. the number of parameters. In this paper, we propose a simple way to improve the capacity of any CNN model having large-scale features, without adding more parameters. In particular, we modify a standard convolutional layer to have a new functionality of channel-selectivity, so that the layer is trained to select important channels to re-distribute their parameters. Our experimental results under various CNN architectures and datasets demonstrate that the proposed new convolutional layer allows new optima that generalize better via efficient resource utilization, compared to the baseline.

논문제목 : Deep Metric Learning Beyond Binary Supervision

저자 : Sungyeon Kim, Minkyo Seo(POSTECH), Ivan Laptev(Inria), Minsu Cho, Suha Kwak(POSTECH)

학술대회 : CVPR 2019

발표자 : 김성연(포항공대)

지도교수 : 곽수하

논문초록 : Metric Learning for visual similarity has mostly adopted binary supervision indicating whether a pair of images are of the same class or not. Such a binary indicator covers only a limited subset of image relations, and is not sufficient to represent semantic similarity between images described by continuous and/or structured labels such as object poses, image captions, and scene graphs. Motivated by this, we present a novel method for deep metric learning using continuous labels. First, we propose a new triplet loss that allows distance ratios in the label space to be preserved in the learned metric space. The proposed loss thus enables our model to learn the degree of similarity rather than just the order. Furthermore, we design a triplet mining strategy adapted to metric learning with continuous labels. We address three different image retrieval tasks with continuous labels in terms of human poses, room layouts and image captions, and demonstrate the superior performance of our approach compared to previous methods.

논문제목 : FickleNet: Weakly and Semi-supervised Semantic Image Segmentation using Stochastic Inference

저자 : Jungbeom Lee, Eunji Kim, Sungmin Lee, Jangho Lee, Sungroh Yoon(Seoul National University)

학술대회 : CVPR 2019

발표자 : 이정범(서울대)

지도교수 : 윤성로

논문초록 : The main obstacle to weakly supervised semantic image segmentation is the difficulty of obtaining pixel-level information from coarse image-level annotations. Most methods based on image-level annotations use localization maps obtained from the classifier, but these only focus on the small discriminative parts of objects and do not capture precise boundaries. FickleNet explores diverse combinations of locations on feature maps created by generic deep neural networks. It selects hidden units randomly and then uses them to obtain activation scores for image classification. FickleNet implicitly learns the coherence of each location in the feature maps, resulting in a localization map which identifies both discriminative and other parts of objects. The ensemble effects are obtained from a single network by selecting random hidden unit pairs, which means that a variety of localization maps are generated from a single image. Our approach does not require any additional training steps and only adds a simple layer to a standard convolutional neural network; nevertheless it outperforms recent comparable techniques on the Pascal VOC 2012 benchmark in both weakly and semi-supervised settings.

[컴퓨터시스템]

논문제목 : TeksDB: Weaving Data Structures for a High-Performance Key-Value Store

저자 : Youil Han(Chungbuk National University), Bryan S. Kim(Seoul National University), Jeseong Yeon(Chungbuk National University), Sungjin Lee(DGIST), Eunji Lee(Soongsil University)

학술대회 : ACM SIGMETRICS 2019

발표자 : 한유일(숭실대)

지도교수 : 이은지

논문초록 : In this paper, we examine the design tradeoffs of existing in-memory data structures of a state-of-the-art key-value store. We observe that no data structures provide both fast point-accesses and consistent ranged- retrievals, and nai?ve amalgamations of existing structures fail to get the best of both worlds. Furthermore, our experiments reveal a performance anomaly when increasing the memory size: as more key-value pairs are maintained in memory, the shortcomings of the data structures exacerbate. To address the above problems, we present TeksDB, a fast and consistent key-value store with a novel in-memory data structure, which efficiently handles both point- and ranged- accesses at a modest increase in memory footprint. Our evaluation demonstrates that TeksDB outperforms RocksDB by 3.6×, 9×, and 4.5× for get, scan, and range_query, respectively. The effectiveness of TeksDB extends to real-world workloads, achieving up to 3.3× speedup for YCSB.

논문제목 : Amoeba: An Autonomous Backup and Recovery SSD for Ransomware Attack Defense

저자 : Donghyun Min, Donggyu Park, Jinwoo Ahn(Sogang University), Ryan Walker, Junghee Lee(University of Texas at San Antonio), Sungyong Park, Youngjae Kim(Sogang University)

학술대회 : HPCA 2019

발표자 : 민동현(서강대)

지도교수 : 김영재

논문초록 : Ransomware is one of growing concerns in personal computer, enterprise and government organizations, because it causes financial damages or loss of important data. Although there are techniques to detect and prevent ransomware, an evolved ransomware may evade them because they are based on monitoring known behaviors. Ransomware can be mitigated if backup copies of data are retained in a safe place. However, existing backup solutions may be under ransomware’s control and an intelligent ransomware may destroy backup copies too. They also incur overhead to storage space, performance and network traffic(in case of remote backup). In this paper, we propose an SSD system that supports automated backup, called Amoeba. In particular, Amoeba is armed with a hardware accelerator that can detect the infection of pages by ransomware attacks at high speed and a fine-grained backup control mechanism to minimize space overhead for original data backup. For evaluation, we extended the Microsoft SSD simulator to implement Amoeba and evaluated it using the realistic block-level traces, which are collected while running the actual ransomware. According to our experiments, Amoeba has negligible overhead and outperforms in performance and space efficiency over the state-of-the-art research, FlashGuard, which supports data backup within the device.

논문제목 : Design Tradeoffs for SSD Reliability

저자 : Bryan S. Kim(Seoul National University), Jongmoo Choi(Dankook University), Sang Lyul Min(Seoul National University)

학술대회 : FAST 2019

발표자 : 김석준(서울대)

지도교수 : 민상렬

논문초록 : Flash memory-based SSDs are popular across a wide range of data storage markets, while the underlying storage medium-;flash memory-;is becoming increasingly unreliable. As a result, modern SSDs employ a number of in-device reliability enhancement techniques, but none of them offers a one size fits all solution when considering the multi-dimensional requirements for SSDs: performance, reliability, and lifetime. In this paper, we examine the design tradeoffs of existing reliability enhancement techniques such as data re-read, intra-SSD redundancy, and data scrubbing. We observe that an uncoordinated use of these techniques adversely affects the performance of the SSD, and careful management of the techniques is necessary for a graceful performance degradation while maintaining a high reliability standard. To that end, we propose a holistic reliability management scheme that selectively employs redundancy, conditionally re-reads, judiciously selects data to scrub. We demonstrate the effectiveness of our scheme by evaluating it across a set of I/O workloads and SSDs wear states.

논문제목 : Graphics-aware Power Governing for Mobile Devices

저자 : Yonghun Choi, Seonghoon Park, Hojung Cha(Yonsei University)

학술대회 : MobiSys 2019

발표자 : 최용훈(연세대)

지도교수 : 차호정

논문초록 : Graphics increasingly play a key role in modern mobile devices. The graphics pipeline requires a close relationship between the CPU and the GPU to ensure energy efficiency and the user’s quality of experience(QoE). Our preliminary analysis showed that the current techniques employed to achieve energy efficiency in the Android graphics pipeline are not optimized especially in the frame generation process. In this paper, we aim to improve the energy efficiency of the Android graphics pipeline without degrading the user’s QoE. To achieve this goal, we studied the internals of the Android graphics pipeline and observed the energy inefficiency in the existing governing framework of the CPU and GPU. Based on the findings, we propose three techniques for addressing energy inefficiency:(1) aggressively capping the maximum CPU frequency,(2) lowering the CPU frequency by raising the GPU minimum frequency, and(3) allocating the frame rendering?related threads in the energy-efficient CPU cores. These techniques are integrated into a single governing framework, called the GFX Governor, and implemented in the newest Android-based smartphones. Experimental results show that without hampering the user’s QoE the average energy consumption of Nexus 6P, Pixel XL, and Pixel 2 XL is reduced at the device level by 24.2%, 18.6%, and 13.7%, respectively, for the 60 chosen applications. We also analyzed the efficacy of the proposed technique in comparison with the state-of-the-art Energy-Aware Scheduling(EAS) implemented in the latest smartphone.

논문제목 : Fast and Efficient Information Transmission with Burst Spikes in Deep Spiking Neural Networks

저자 : Seongsik Park, Seijoon Kim, Hyeokjun Choe, Sungroh Yoon(Seoul National University)

학술대회 : DAC 2019

발표자 : 박성식(서울대)

지도교수 : 윤성로

논문초록 : Spiking neural networks (SNNs) are considered as one of the most promising artificial neural networks due to their energy-efficient computing capability. Recently, conversion of a trained deep neural network to an SNN has improved the accuracy of deep SNNs. However, most of the previous studies have not achieved satisfactory results in terms of inference speed and energy efficiency. In this paper, we propose a fast and energy-efficient information transmission method with burst spikes and hybrid neural coding scheme in deep SNNs. Our experimental results showed the proposed methods can improve inference energy efficiency and shorten the latency.

논문제목 : CIDR: A Cost-Effective In-line Data Reduction System for Terabit-per-Second Scale SSD Arrays

저자 : Mohammadamin Ajdari(POSTECH), Pyeongsu Park, Joonsung Kim, Dongup Kwon, Jangwoo Kim(Seoul National University)

학술대회 : HPCA 2019

발표자 : Mohammadamin Ajdari(포항공대)

지도교수 : 김장우

논문초록 : An SSD array, a storage system consisting of multiple SSDs per node, has become a design choice to implement a fast primary storage system, and modern storage architects now aim to achieve terabit-per-second scale performance with the next-generation SSD array. To reduce the storage cost and improve the device endurability, such SSD array must employ data reduction schemes(i.e., deduplication, compression), which provide high data reduction capability at minimum costs. However, existing data reduction schemes do not scale with the fast increasing performance of an SSD array, due to inhibitive amount of CPU resources(e.g., in software-based schemes) or low data reduction ratio(e.g., in SSD device wide deduplication) or being cost ineffective to address workload changes in datacenters(e.g., in ASIC-based acceleration). In this paper, we propose CIDR, a novel FPGA-based, cost-effective data reduction system for an SSD array to achieve the terabit-per-second scale storage performance. Our key ideas are as follows. First, we decouple data reduction-related computing tasks from the unscalable host CPUs by offloading them to a scalable array of FPGA boards. Second, we employ a centralized, node-wide metadata management scheme to achieve an SSD array-wide, high data reduction. Third, our FPGA-based reconfiguration adapts to different workload patterns by dynamically balancing the amount of software and hardware tasks running on CPUs and FPGAs, respectively. For evaluation, we built our example CIDR prototype achieving up to 12.8 GB/s(0.1 Tbps) on one FPGA. CIDR outperforms the baseline for a write-only workload by up to 2.47x and a mixed read-write workload by an expected 3.2x, respectively. We showed CIDR’s scalability to achieve Tbps-scale performance by measuring a two-FPGA CIDR and projecting the performance impacts for more FPGAs.

논문제목 : LpGL: Low-power Graphics Library for Mobile AR Headsets

저자 : Jaewon Choi, HyeonJung Park(Ajou University), Jeongyeup Paek(Chung-Ang University), Rajesh Krishna Balan(Singapore Management University), JeongGil Ko(Ajou University)

학술대회 : MobiSys 2019

발표자 : 최재원(아주대)

지도교수 : 고정길

논문초록 : We present LpGL, an OpenGL API compatible Low-power Graphics Library for energy efficient AR headset applications. We first characterize the power consumption patterns of a state of the art AR headset, Magic Leap One, and empirically show that its internal GPU is the most impactful and controllable energy consumer. Based on the preliminary studies, we design LpGL so that it uses the device’s gaze/head orientation information and geometry data to infer user perception information, intercepts application-level graphics API calls, and employs frame rate control, mesh simplification, and culling techniques to enhance energy efficiency of AR headsets without detriment of user experience. Results from a comprehensive set of controlled in-lab experiments and an IRB-approved user study with 25 participants show that LpGL reduces up to ∼22% of total energy usage while adding only 46 μsec of latency per object with close to no loss in subjective user experience.

[컴퓨터이론]

논문제목 : Efficient Subgraph Matching: Harmonizing Dynamic Programming, Adaptive Matching Order, and Failing Set Together

저자 : Myoungji Han, Hyunjoon Kim, Geonmo Gu, Kunsoo Park(Seoul National University), Wook-Shin Han(POSTECH)

학술대회 : SIGMOD 2019

발표자 : 구건모(서울대)

지도교수 : 박근수

논문초록 : Subgraph matching(or subgraph isomorphism) is one of the fundamental problems in graph analysis. Extensive research has been done to develop practical solutions for subgraph matching. The state-of-the-art algorithms such as CFL-Match and TurboISO convert a query graph into a spanning tree for obtaining candidates for each query vertex and obtaining a good matching order with the spanning tree. However, by using the spanning tree instead of the original query graph, it could lead to lower pruning power and a sub-optimal matching order. Another limitation is that they perform redundant computation in search without utilizing the knowledge learned from past computation. In this paper, we introduce three novel concepts to address these inherent limitations: 1) dynamic programming between a directed acyclic graph(DAG) and a graph, 2) adaptive matching order with DAG ordering, and 3) pruning by failing sets, which together lead to a much faster and more robust algorithm DAF for subgraph matching. Extensive experiments with real datasets show that DAF outperforms the fastest existing solution by up to orders of magnitude in terms of recursive calls as well as in terms of elapsed time.


서울시 서초구 방배로 76 (방배동, 머리재빌딩 401호) 우)06704 | (Tel)1588-2728 | (Fax)02-521-1352 | 사업자등록번호 : 114-82-03170 | 대표 : 나연묵

Copyright (c) KIISE. All rights reserved.