KCC 2023 여성위원회 특별 세션

신진 여성 연구자 스포트라이트 세션: 여성정보인 Meet-Up Day

최근에 학위를 받은 국내외의 신진 여성연구인력을 소개하고 최신의 연구 성과를 공유하며, KCC 여성참여자들이 친근한 분위기 속에서 교류할 수 있는 만남의 장을 마련했습니다.

• 일시: 6.19(월) 09:30-12:00

• 장소: 라마다프라다제주호텔 2층 우도홀

• 프로그램

09:30 ~ 09:35 개회사 

임은진 여성위원장 (사회자)

09:35 ~ 09:40          인사말 

                                이원준 정보과학회장

09:40 ~ 10:00 How to Reconstruct and Analyze Human Movement from Video 

유리 교수(아주대학교)

10:00 ~ 10:20 Systems for Tiny Machine Learning 

허선영 교수(경희대학교)

10:20 ~ 10:40 Cryptographic Techniques for Privacy-Preserving Computation

서민혜 교수(덕성여자대학교)

10:40 ~ 10:50 Break

10:50 ~ 11:10 Parameterized Algorithms for the Planar Disjoint Paths Problem 

오은진 교수(포항공과대학교)

11:10 ~ 11:30 Adversarial understanding on anonymous network systems using deep learning

오세은 교수(이화여자대학교)

11:30 ~ 11:40 Closing

발표 상세 소개

유리 교수 (아주대학교),  How to Reconstruct and Analyze Human Movement from Video

Reconstructing and analyzing human movement are interesting but challenging research topics in many fields such as computer graphics, computer vision, biomechanics, and robotics. To obtain human motion, motion capture system or sensor-based methods have been used as a gold standard. However, these sensor-based approaches are costly and time-consuming.

2D video data is a good source for extracting human movement, and easy to get with a smartphone camera. These days, thanks to the remarkable development of deep learning, human pose estimation techniques from video show reasonable performance. However, 2D video is lack of information inherently, thus poses estimated from video suffer from pose ambiguities and occlusion issues. Our research proved that by using physics simulation considering interactions with the environment can tackle this problem and reconstruct human motion successfully. Research reconstructing human motion from 2D video in 3D virtual space can help analyzing and understanding human more easily and conveniently.

In this talk, we explain how to reconstruct and control human motion from 2D video using pose estimation techniques and deep reinforcement learning. Also, we introduce the “gait analysis tool” which takes gait video as an input and outputs gait parameters for Parkinson’s disease patients.

허선영 교수 (경희대학교), Systems for Tiny Machine Learning  

Tiny machine learning is to enable machine learning applications on small embedded systems, which normally employ low-power microcontrollers. Deploying machine learning models is challenging for small embedded systems because they have too limited resources to meet system requirements (e.g., real-timeness). Especially, the main memory of less than one megabyte is considered as the main limiting factor in executing machine learning models on those systems. This talk will discuss how to optimize machine learning models for small embedded systems to enhance memory efficiency and reduce execution time.

서민혜 교수 (덕성여자대학교), Cryptographic Techniques for Privacy-Preserving Computation 

abstract 

오은진 교수 (POSTECH), Parameterized Algorithms for the Planar Disjoint Paths Problem 

Given a planar graph G with n  vertices and a set T= {(s_1,t_1) ... (s_k,t_k)}  of k terminal pairs, the disjoint paths problem asks for computing a set of vertex-disjoint paths  p_1,...p_k such that  p_i  connects. s_i  and t_j . In this talk, I will introduce a  $2^{O(k^2)}n$-time algorithm for the planar disjoint paths problem. This improves the previously best-known algorithms with running times of $2^{O(k^2)}n^6$ and $2^{2^{O(k)}}n$.

오세은 교수 (이화여자대학교), Adversarial understanding on anonymous network systems using deep learning

Tor is one of the most widely used anonymous network systems with millions of daily users. It consists of three hops to relay the traffic between the client and destination to hide both ends of the communication. Despite such wide adoption, the extent to which anonymity is guaranteed through Tor is still questionable. In this talk, we introduce two types of potential attacks on Tor; website fingerprinting and end-to-end flow correlation. This research urges Tor developers to implement traffic analysis-resistant defenses.

[최종 수정일: 2023.05.12]