user_image

Changyeol Lee

Graduate student studying Computer Science at Yonsei Univ.
Combinatorial Optimization Lab (Advisor - Prof. Hyung-Chan An)
Email - changyeollee_at_yonsei.ac.kr
#Social #Creative #Positive

EDUCATION

  • Ph.D student at Yonsei Univserity (advised by Prof. Hyung-Chan An)
  • B.S. in Computer Science, Yonsei Univ. (2017.03-2021.02)

RESEARCH INTERESTS

  • Approximation Algorithm
  • Online Algortihm
  • Learning-augmented Algorithm
  • Combinatorial Optimization
  • Miscellaneous (Theoretical Computer Science, Game Theory and Mechanism Design, Quantum algorithms)

AWARDS

  • B.S. in Computer Science, Yonsei Univ. High honors at graduation (2021.02)
  • RC Creative Platform Grand Prize, Yonsei Univ. & NEXON Co., Ltd. (2017.11)

EXPERIENCE

  • 2024.03. – current

    Co-organizer of Yonsei CS Theory Student Group(link)
    (with Sungmin Kim(link))

  • 2022.09. – 2022.10.

    Visiting Research Intern, Cornell Univ. (Advisor - Prof. David Shmoys)

  • 2021.03. – 2024.02.

    Teaching Assistent, Yonsei Univ. (Data Structures, Algorithm Analysis)

  • 2019.03. – 2020.12.

    Tutoring, Yonsei Univ. (Linear Algebra, Algorithm Analysis)

  • 2018.12. – 2021.02.

    Research Intern, Yonsei Univ. (Advisor - Prof. Hyung-Chan An)

  • 2018.05. – 2020.12.

    Yonsei-Nexon RC Creative Platform Peer Mentoring (Head Mentor)

  • 2017.03. – 2019.02.

    Student Council of Computer Science Department, Yonsei Univ.

PROJECTS

few shot learning

Fall 2019

A Few-Shot Learning for Animation Character Classification
Few-shot classification is a problem of recognizing unseen classes during training with limited labeled examples. While meta-learning algorithms became one promising direction, our experiment shows that these algorithms work poorly on animated character datasets compared to real-world image datasets. To overcome this problem, we propose two approaches, 1) introducing triple loss (or quadruple loss) to the existing algorithms, and 2) converting features to be more task-relevant using learnable mapping function.
Positive and Negative Training of Graph Convolutional Network

Spring 2020

Positive and Negative Training of Graph Convolutional Network
Graph convolutional network (GCN) is a class of convolutional network that can process data that can be represented by graph. Utilizing that the concept of negativity is clear on graph, we propose the training paradigm that considering both positive graph and negative graph where negative graph can be defined various way, e.g., shortest path, random drop. The experiment shows that applying this training to GCN models (e.g., GraphSAGE, GAT, SplinCNN) can outperform the existing accuracy.

RECENT PUBLICATIONS

[All Publications]
  1. On Optimal Consistency-Robustness Trade-Off for Learning-Augmented Multi-Option Ski Rental
    Shin, YonghoLee, Changyeol, and An, Hyung-Chan
    arXiv preprint arXiv:2312.02547, 2023
  2. Improved Learning-Augmented Algorithms for the Multi-Option Ski Rental Problem via Best-Possible Competitive Analysis
    Shin, YonghoLee, Changyeol, Lee, Gukryeol, and An, Hyung-Chan
    In Proceedings of the 40th International Conference on Machine Learning, vol. 202, pp. 31539–31561, 23–29 jul, 2023