특별 프로그램

발표 제목 : 미디어와 인공지능

Abstract
대중매체인 방송이나 언론으로 인식되던 ‘미디어’가 유무선 인터넷의 발달로 인해 그 형태가 많이 다양해지고 있으며, 최근 회자되고 있는 메타버스는 차세대 인터넷으로 여겨질 뿐 아니라 새로운 소통수단으로서의 미디어로 생각할 수 있을 것이다. 이렇게 다양한 형태를 가지는 미디어에 대해 그 의미를 다시 한 번 생각해보고, media for AI, media by AI 등 미디어 영역에서의 인공지능의 활용에 대해 알아본다.

약력
– 1993. 1. ~ 2001.12.: POSCO 기술연구소 선임연구원
– 2002. 1. ~ 2003. 10.: (주)맥스웨이브 연구개발팀 팀장
– 2004.2. ~ 현재: 한국전자통신연구원 근무 (현, 미디어연구본부 본부장)
– 현, 과학기술정보통신부 미디어 자문위원(2021.2. ~ )
– 현, 한국방송미디어공학회 협동부회장
– 현, 미래방송미디어표준포럼 운영위원

발표 제목 : DRANet: Disentangling Representation and Adaptation Networks for Unsupervised Cross-Domain Adaptation, CVPR2021

Abstract
Deep neural networks (DNNs) have led to significant performance improvements in a variety of areas but can fail badly on real-world images outside the dataset. In this talk, I present DRANet that tackles the issue by disentangling image representations and transferring the visual attributes in a latent space. Unlike the existing domain adaptation methods that learn associated features sharing a domain, DRANet preserves the distinctiveness of each domain’s characteristics and allows bi-/multi-directional domain adaptation with a single encoder-decoder network. This work will be published in CVPR2021.

약력
– Assistant Professor, DGIST, Daegu, Korea, 2019.09-present
– Visiting Scholar, Carnegie Mellon University, PA, USA, 2019.06-2019.08
– Research Intern, Microsoft Research, Beijing, China, 2018.02-2018.08
– Ph.D., KAIST, 2016.03-2019.08
– Microsoft Research Fellowship, 2018.
– Global Ph.D. Fellowship, 2016.

발표 제목 :
– SVIAS: Scene-segmented Video Annotation System Ver 1.0 and 2.0
– ACM International Conference on Multimedia, 2018 and 2020.

Abstract
We had designed the scene-segmented video information annotation system using video segmentation and information annotation for several years. For video segmentation, the proposed system adapts the multiview deep convolution neural network in version 1.0 and upgrades it with a scene context in version 2.0. Segmented scenes are annotated by using the supervised movie captioning model. In the movie captioning model, the structure of movie scripts is reflected into the model to generate script-like sentences. Both functionalities effectively work together with the web interface designed to tie our functionalities and external content providers.

약력
– Senior researcher in Smart media research group, ETRI, 2013. 9. ~
– Ph. D. in Computer Science, Kyungpook National University, 2012. 2.
– M.S. in Computer Science, Kyungpook National University, 2007. 2.
– B.S. in Computer Science, Kyungpook National University, 2005. 2.

발표 제목 : GAN 기술의 현황과 그 활용

Abstract
본 튜토리얼에서는 최근 적대적생성모델(GAN)과 그 응용분야에 대하여 소개한다. GAN은 생성모듈과 판별모듈이 서로 경쟁적으로 훈련되는 방식에 기반을 둔 심층생성모델로, 처음 발명된 이후 가짜 이미지 생성과 스타일 변환 등에 주로 이용되었다. 현재 GAN의 응용범위는 시계열 데이터와 같은 정형데이터로도 확장되고 있다. 본 튜토리얼은 먼저 GAN에 대한 일반적인 내용을 소개하고, 주어진 조건에 따라 가짜 데이터를 생성하는 조건부 GAN에 대하여 설명한다. 그리고 분류모듈을 따로 갖고 있는 조절가능한 GAN에 대해서 설명하고, GAN을 시계열 데이터에 적용하는 방법에 대해서도 이야기한다. 마지막으로 금융데이터에 GAN을 적용한 케이스 연구에 대해서도 이야기한다.

약력
– 2014 ~ 현재, 고려대학교 전기전자공학부 부교수
– 2013 ~ 2014, 미국 노스웨스턴 대학교, 의생명정보학, 조교수
– 2011 ~ 2013, 미국 스탠포드 대학교, 통계학과, 박사후 연구원
– 2011, 미국 스탠포드 대학교, 전기전자공학 박사

발표 제목 : 다중 관계 그래프를 이용한 보행자 이동 경로 예측

Abstract
보행자 이동 경로 예측은 자율주행과 소셜 로보틱스 분야의 핵심 기능 중 하나로써 각광을 받고 있다. 보행자 이동경로 예측은 일정 시간동안 보행자의 이동경로를 바탕으로 미래의 이동 경로를 예측하는 것으로써, 자율주행 자동차와 모바일 로봇이 보행자에게 심리적으로 허용가능한 회피/주행을 구현하는 데 필요한 기능으로 기대를 모으고 있다. 2015년 인공지능 기법이 컴퓨터비전 분야에 활발히 적용되기 시작하면서, Stanford University, UC Berkeley, Carnegie Mellon University 등 해외의 주요 연구 그룹에서는 보행자의 이동 경로 예측 기법을 고도화 시키고 그 실용적 가능성을 높이고 있다.
본 세미나에서 보행자 이동경로 예측의 세계적 기술 동향을 소개하고, 본 연사의 최근 관련 연구 실적을 소개한다. 특히, 본 연사의 연구팀은 세계 최고 수준의 예측 정확도를 갖는 그래프 뉴럴 네트워크 기반의 보행자 이동경로 추정 방법을 세계 최고의 인공지능 학회인 AAAI2021에 발표하였다. 영상 속 보행자들간의 다관계성을 그래프 형식으로 모델링하고, 이에 대한 최적 학습기법을 제안하였다. 제안된 알고리즘은 기존 알고리즘 대비 약 30%의 성능향상을 보였고, 그 우수성을 인정받아 2021 휴먼테크 논문대상 컴퓨터과학분과 동상을 수상하였다.

약력
– Assistant Professor, GIST, AI Graduate School
– KAIST, Daejeon, Korea, Ph.D, Electrical Engineering, Mar 2015 – Feb 2018
– KAIST, Daejeon, Korea, M.S., Electrical Engineering, Aug 2011 – Aug 2013
– Yonsei University, Seoul, Korea, B.S., Electrical and Electronic Engineering, Aug 2011

발표 제목 : Understanding the Potential Risks of Sharing Elevation Information on Fitness Applications

Abstract
The extensive use of smartphones and wearable devices has facilitated many useful applications. For example, with Global Positioning System (GPS)-equipped smart and wearable devices, many applications can gather, process, and share rich metadata, such as geolocation, trajectories, elevation, and time. For example, fitness applications, such as Runkeeper and Strava, utilize information for activity tracking, and have recently witnessed a boom in popularity. Those fitness tracker applications have their own web platforms, and allow users to share activities on such platforms, or even with other social network platforms. To preserve privacy of users while allowing sharing, several of those platforms may allow users to disclose partial information, such as the elevation profile for an activity, which supposedly would not leak the location of the users. In this work, and as a cautionary tale, we create a proof of concept where we examine the extent to which elevation profiles can be used to predict the location of users. To tackle this problem, we devise three plausible threat settings under which the city or borough of the targets can be predicted. Those threat settings define the amount of information available to the adversary to launch the prediction attacks. Establishing that simple features of elevation profiles, e.g., spectral features, are insufficient, we devise both natural language processing (NLP)-inspired text-like representation and computer vision-inspired image-like representation of elevation profiles, and we convert the problem at hand into text and image classification problem. We use both traditional machine learning-and deep learning-based techniques, and achieve a prediction success rate ranging from 59.59% to 95.83%. The findings are alarming, and highlight that sharing elevation information may have significant location privacy risks.

약력
– Korea University, B.S. in Computer Science and Engineering (2004)
– Korea University, M.S. in Computer Science and Engineering (2006)
– University of Southern California, Ph.D. in Computer Science (2014)
– Intel Corporation, Systems Engineer (2013-2016)
– Chung-Ang University, Assistant Professor (2016-2019)
– Korea University, Associate Professor (2019-Present)

발표 제목 : Learning-based DVFS with Zero Thermal Throttling for Mobile Devices

Abstract
DVFS (dynamic voltage and frequency scaling) is a system-level technique that adjusts voltage and frequency levels of CPU/GPU at runtime to balance energy efficiency and high performance. DVFS has been studied for many years, but it is considered still challenging to realize a DVFS that performs ideally for mobile devices for two main reasons: i) an optimal power budget distribution between CPU and GPU in a power-constrained platform can only be defined by the application performance, but conventional DVFS implementations are mostly application-agnostic; ii) mobile platforms experience dynamic thermal environments for many reasons such as mobility and holding methods, but conventional implementations are not adaptive enough to such environmental changes. In this work, we propose a deep reinforcement learning-based frequency scaling technique, zTT. zTT learns thermal environmental characteristics and jointly scales CPU and GPU frequencies to maximize the application performance in an energy-efficient manner while achieving zero thermal throttling. Our evaluations for zTT implemented on Google Pixel 3a and NVIDIA JETSON TX2 platform with various applications show that zTT can adapt quickly to changing thermal environments, consistently resulting in high application performance with energy efficiency.

약력
– Associate Professor, Seoul National University, the Department of Electrical and Computer Engineering
– KAIST, B.S. in Electrical Engineering,
– KAIST, M.S. in Electrical Engineering
-KAIST, Ph. D. in Electrical Engineering

-He is serving as an Editor for IEEE/ACM Transactions on Networking, IEEE Transactions on Vehicular Technology, Computer Networks (Elsevier), and ICT Express. He is also serving as a General co-Chair of ACM MobiHoc 2022. He received two IEEE William R. Bennett Prizes from IEEE ComSoc in 2013 and 2016, respectively, and received the best paper award from ACM MobiSys 2021. His research interests include low-latency and performance guaranteed networking for 5G/6G/DataCenters, mobile machine learning, learning-based image/video/data encoding, offloaded neural network processing, low-power computing, mobile systems for augmented/mixed reality, and human context modeling.

발표 제목 : Advisable Learning for Self-driving Vehicles by Internalizing Observation-Action Rules / CVPR 2020

Abstract
Recent success suggests that deep neural control networks are likely to be a key component of self-driving vehicles. These networks are trained on large datasets to imitate human actions, but they lack semantic understanding of image contents. This makes them brittle and potentially unsafe in situations that do not match training data. Here, we propose to address this issue by augmenting training data with natural language advice from a human. Advice includes guidance about what to do and where to attend. We present the first step toward advice-giving, where we train an end-to-end vehicle controller that accepts advice. The controller adapts the way it attends to the scene (visual attention) and the control (steering and speed).

약력
학력:
– 2008 고려대 전기전자공학 학사
– 2010 고려대 전자컴퓨터공학 석사
– 2019 UC Berkeley 컴퓨터과학 박사

경력:
– 2010.10 – 2014.3 LG Display 연구센터 주임연구원
– 2020.1 – 2021.2 Waymo Research Scientist

발표 제목 : Multi-Sample Online Learning for Spiking Neural Networks based on Generalized Expectation Maximization / ICASSP 2021

Abstract
Spiking Neural Networks (SNNs) offer a novel computational paradigm that captures some of the efficiency of biological brains by processing through binary neural dynamic activations. Probabilistic SNN models are typically trained to maximize the likelihood of the desired outputs by using unbiased estimates of the log-likelihood gradients. While prior work used single-sample estimators obtained from a single run of the network, this paper proposes to leverage multiple compartments that sample independent spiking signals while sharing synaptic weights. The key idea is to use these signals to obtain more accurate statistical estimates of the log-likelihood training criterion, as well as of its gradient. The approach is based on generalized expectation-maximization (GEM), which optimizes a tighter approximation of the log-likelihood using importance sampling. The derived online learning algorithm implements a three-factor rule with global per-compartment learning signals. Experimental results on a classification task on the neuromorphic MNIST-DVS data set demonstrate significant improvements in terms of log-likelihood, accuracy, and calibration when increasing the number of compartments used for training and inference.

약력
– 동국대학교 인공지능학과 조교수 (2021-현재)
– King’s College London 박사후연구원 (2018-2021)
– KAIST 전기 및 전자공학과 박사 (2017)
– KAIST 전기 및 전자공학과 석사 (2012)
– KAIST 전기 및 전자공학과 학사 (2010)

발표 제목 : A gentle introduction to Generative Adversarial Network (GAN)

Abstract
Generative adversarial networks (GANs) have evolved dramatically in recent years, enabling high-fidelity image synthesis with models learned directly from data. Not only limited to random image generation, but GAN has been heavily explored in various topics such as image-to-image translation, disentangled manipulation, and real image editing. In this tutorial, we will present an introduction and trends in GAN.

약력
– 2019 서울대학교 컴퓨터공학부 졸업(학사)
– 2021 서울대학교 컴퓨터공학부 졸업(석사)
– 2016~2017 코드윙즈, CEO
– 2017 Candy Camera, Server Engineer
– 2021~현재 NAVER AI Lab, Research Scientist