ACADEMIA
RESEARCH EXPERIENCES
GRADUATE STUDENT (MARCH 2022 ~ )
Brain and Machine Intelligence Lab. (BMIL)
Department of Electronic Engineering, Soongsil University, Seoul, Republic of Korea
석사 과정(MS) - 2022년 3월 ~
숭실대학교 정보통신공학부, BMI Lab.
UNDERGRADUATE RESEARCHER (MARCH 2021 ~ FEBRUARY 2022)
Brain and Machine Intelligence Lab. (BMIL)
Department of Electronic Engineering, Soongsil University, Seoul, Republic of Korea
학부 연구원 (RA) - 2021년 3월 ~ 2022년 2월
숭실대학교 정보통신전자공학부, BMI Lab.
UNDERGRADUATE RESEARCH ASSISTANT (DECEMBER 2020 ~ FEBRUARY 2021 )
Brain and Machine Intelligence Lab. (BMIL)
Department of Electronic Engineering, Soongsil University, Seoul, Republic of Korea
학부 연구원 보조 (URA) - 2020년 12월 ~ 2021년 2월
숭실대학교 정보통신전자공학부, BMI Lab.
INTERNATIONAL PAPER
[AI-L1] Dongsu Lee, Minhae Kwon, "ADAS-RL: Safety Learning Approach for Stable Autonomous Driving," ICT express, Accepted. (IF 4.317, Rank 22/91, Top 23.26%)
Stability is the most significant component of an autonomous driving system, affecting both the lives of drivers and pedestrians and traffic flow. Reinforcement learning (RL) is a representative technology used in autonomous driving, but it has challenges because it is based on trial and error. In this letter, we propose an efficient learning approach for stable autonomous driving. The proposed deep reinforcement learning based approach can address the partially observable scenario in mixed traffic which includes both autonomous vehicles and human-driven vehicles. Simulation results show that the proposed model outperforms the control-theoretic and vanilla RL approaches. Furthermore, we confirm the effect of the sync-penalty, which teaches the agent about unsafe decisions without experiencing the accidents.
[AI-W1] Dongsu Lee, Minhae Kwon, “Stability Analysis in Mixed-Autonomous Traffic with Deep Reinforcement Learning,” Conference on Neural Information Processing Systems (NeurIPS) Deep Reinforcement Learning (DeepRL) Workshop, December 2021. [Media]
With the development of deep neural networks and artificial intelligence, Autonomous Driving Systems (ADS) are developing rapidly. According to the commercialization of Autonomous Vehicles (AVs), non-AVs and AVs will drive simultaneously on the road. The stability of autonomous vehicles can significantly affect the entire road condition. In this study, we use a Deep Reinforcement Learning (DRL) approach to making an AV learn a reasonable lane-changing and the acceleration control to keep the desired velocity. For the learning efficiency of the AV, it provides minimal state information and replaces the lane-changing action space with a lower level. Therefore, we modified the action selection method of TD3 and used it. Finally, the driving performance of the TD3-based AV and the LC2013-based vehicle is compared in various environments. The TD3-based AV performed better than the LC 2013.
DOMESTIC JOURNAL ARTICLE
이동수, 권민혜, “심층강화학습기반 자율주행차량을 이용한 원형도로의 STOP-AND-GO WAVE 현상 해결 전략 연구,” 한국통신학회 논문지, VOL.46, NO.10, 2021년 10월.
Combating Stop-and-Go Wave Problem at a Ring Road Using Deep Reinforcement Learning Based Autonomous Vehicles
With the rapid development of artificial intelligence, autonomous driving has recently attracted considerable attention. This paper aims to use an autonomous vehicle to improve road flow by solving the stop-and-go-wave problem on a ring road. We design a special model of Markov decision process model to solve stop-and-go wave and use three deep reinforcement learning algorithms to train autonomous vehicles: Proximal Policy Optimization (PPO), Deep Deterministic Policy Gradient (DDPG), and Twin Delayed DDPG (TD3). We then compare their driving patterns and performances. We confirmed that an autonomous vehicle on the ring road could control the flow of multiple non-autonomous vehicles with an extensive simulation study, thus successfully solving the stop-and-go wave problem.
TEACHING ASSISTANT
PROGRAMMING AND PRACTICE, WINTER, 2021.
DOMESTIC CONFERENCE
이동수, 권민혜, "다중 개체 간 네트워크 공유를 통한 심층강화학습의 학습 효율성 향상 연구," 한국통신학회 동계종합학술대회, FEBRUARY 2022.
Enhancing Training Efficiency of Multi-agent Deep Reinforcement Learning by Sharing Networks
이동수, 권민혜, “심층강화학습 기반 자율주행차량의 차선 변경 정책 안정성 평가,” 한국인공지능학술대회, SEPTEMBER 2021.
Evaluation of Policy Stability in Lane Change based on Deep Reinforcement Learning
이동수, 권민혜, “PPO기반 자율주행차량의 효율적이고 안전한 차선 변경 정책 연구,” 한국통신학회 하계종합학술대회, JUNE 2021.
PPO-based Efficient and Safe Lane Change Strategy for Autonomous Vehicles
이동수, 김선웅, 권민혜, “심층 강화학습기반 자율주행차의 에너지 효율적인 제어 방법 연구,” 통신정보 합동학술대회(JCCI), APRIL 2021.
Deep Reinforcement Learning Based Energy-efficient Control for Autonomous Vehicles
EDUCATION
Undergraduate B.S., Bio-Medical System
Soongsil University, Seoul, Republic of Korea
Current GPA: 4.22/4.5 (8 Semester Avg.)
Transformed GPA: 3.75/4.0 (8 Semester Avg.)
Summa Cum Laude (1/58)
숭실대학교
의생명시스템학부 학사과정
현재 학점 4.22/4.5 (8학기 평균)
환산 학점 3.75/4.0 (8학기 평균)
PROJECTS
With respect to Biology
NETWORK ANALYSIS FOR SINGLE-CELL RNA SEQUENCING DATA OF PLASMODIUM FALCIPARUM
2021.03~2021.06
Malaria is a very serious problem in terms of public health. Specifically, Plasmodium Falciparum is the cause of numerous child deaths every year. In order to prevent and eradicate specific diseases, it is absolutely necessary to understand the life circle at the genetic level. To this end, analysis at the genetic level is performed through scRNA-seq. In this paper, we perform gene co-expression network and Protein-Protein interaction network analysis using scRNA-seq data of Plasmodium Falciparum species. Through network analysis, we provide clues that will accelerate research for the biological understanding and interpretation of Plasmodium Falciparum species. In addition, we analyze major genes and their respective functions, and present major genes that can be considered as targets for disease treatment drugs.
PREDICTION OF SARS-COV-2 MAIN PROTEASE BINDING FREE ENERGY
USING GRAPH CONVOLUTIONAL NETWORKS
2021.03~2021.06
약물과 표적 간의 상호 작용에 대한 계산적 예측은 SARS-CoV-2 (COVID-19)의 신약 개발 과정에 있어 중요한 단계이며, 컴퓨터를 이용하면 이 과정을 가속화하고 자원을 줄일 수 있다. 이를 위한 Drug–Target Binding Affinity (DTA) 예측에 딥 러닝을 도입하고 정확도를 향상시키는 것이 최근 연구의 초점이 되고 있다. 본 논문에서는 분자의 구조 정보를 활용하여 분자 Graph를 생성하고, SARS-CoV-2의 표적 단백질인 Mpro (PDB entry: 6LU7)와 약물 사이의 Binding Free Energy (BFE, kcal/mol) 예측을 위한 Graph Convolutional Networks (GCN)를 제안한다. 분자 Graph는 원소의 화학적 특성을 구조적 특성에 따라 Convolution하여 BFE를 예측할 수 있으며, 모델 개발을 위한 추가적인 복잡한 계산/실험 과정은 필요하지 않다. BFE Dataset는 AutoDock Vina를 이용해 생성된 4,717개 분자의 SMILES를 이용하였다. 결과적으로, 본 논문에서 제안하는 GCN 모델은 BFE 예측을 위한 효과적인 접근 방식이며, 약물 개발 과정에서 매우 유용할 수 있음을 보여준다.
EXPRESSION QUANTITATIVE TRAIT LOCI FOR CCR4-NOT COMPLEX THAT REGULATE GLOBAL GENE EXPRESSION
2020.09~2020.12
Abstract
A genome wide association study(GWAS) was conducted to identify expression quantitative trait loci(eQTLs) for the CCR4-NOT complex that regulated gene expression at all steps. Data derived from RNA expression in lymphoblastoid cells of 373 unrelated Europeans. We analyzed the genetic associations of SNPs with expression of the genes encoding 10 proteins: CNOT1, CNOT2, CNOT3, CNOT4, CCR4a, CAF1, CAF40, CNOT10, CNOT11, and TAB182 among CCR4-NOT complex. In the current study, we revealed 2 eQTLs associated with CNOT4(P < ). One(rs114824303) of them was located the intronic site of the gene encoding WD Repeat Domain 18(WDR18). And then rs114824303 have strong linkage with cis-eQTLs for gene encoding WDR18. WDR18 is well known subunit of Five Friends of Methylated Chromatin Target(5FMC) that regulate gene expression dependent with ZNF148. CNOT4 target Promoter GH07J135507(GeneHancer ID) have ZNF 148 that is one of transcription factor binding sites. The current study suggested 2 novel eQTLs for CNOT4 and association with WDR18 of 5FMC. Further studies are required to understand their underlying mechanisms to unknown pathway that includes CCR4-NOT complex and 5FMC complex.