이준영

이준영 (Joonyoung Yi)

Head of Hyperconnect AI (Jan. 2020 - Present)

LinkedIn | Google Scholar | GitHub

AI/ML 연구로 비즈니스 임팩트를 내는 데 열정을 가지고 있습니다. 현재 Hyperconnect AI를 맡고 있으며, 제품의 추천, Trust & Safety (모더레이션), 그리고 생성기술을 활용한 Innovation을 담당하고 있습니다.

문제를 푸는 것에 관심이 많습니다. 솔루션을 먼저 생각하기보다는 문제를 먼저 잘 정의하고, 그 문제를 가장 잘 풀 수 있는 해결책을 다양한 방법으로 찾고 최선의 방법을 고민합니다. 이렇게 만든 기술을 실제 제품에 적용해보고 결과를 평가합니다. 잘되면 왜 잘되었는지, 부족하면 왜 부족한지를 평가하고 다시 문제 정의로 돌아가는 이터레이션을 돕습니다. 일련의 깊이 있는 문제 풀이(연구) 과정을 통해 비즈니스 임팩트를 내고 있습니다.

Hyperconnect 입사 전에는 두 번의 창업을 경험했고, 안드로이드 개발자로 커리어를 시작했습니다.

Experience

Hyperconnect (part of Match Group)

Director, Head of Hyperconnect AI Aug. 2024 - Present
  • Oversee Hyperconnect's internal AI technologies across recommendation, moderation, and generative AI.
  • Directed Match Group-wide AI moderation system, building scalable infrastructure that safeguards millions of users across Tinder, Azar, and Hinge.
  • Manage 30+ members (ML researchers, software engineers, PMs, and data scientists).
Director, Head of AI Lab Mar. 2024 - Aug. 2024
  • Led key recommendation initiatives across Tinder, Hinge, and Azar.
  • Oversaw early generative AI initiatives, managing the team prior to the spin-off of Mel.
  • Managed 15+ members (ML researchers, software engineers, and data scientists).
ML research team leader Mar. 2023 - Mar. 2024
  • As one of the main contributors, contributed to transforming the Azar matching system from a heuristic-based approach to an AI-based one, resulting in a 24% YoY revenue increase for Azar after years of stagnation following Match Group's acquisition. From Match Group's 2023 Q2 earnings letter, "At Hyperconnect, Azar experienced very strong revenue growth in Q2, largely driven by improvements to its new AI-powered matching algorithm, which is increasing both user engagement and monetization."
  • Participated in the invention and development of a photo selector that recommends suitable profile pictures from the camera roll for use on Tinder and Hinge. Led Hinge's top photo project and improved Hinge's previously stagnant north star metric by 3.6%. From Match Group's 2024 Q2 earnings call, "The Hyperconnect team has been integral in creating several of the AI inhibitors that have been introduced across our brands, including Tinder's Photo Selector and Hinge's Top Photo and Photo Finder".
  • Managed 10+ members (ML researchers and data scientists).
Data science unit lead Mar. 2022 - Mar. 2023
  • Led the data science unit as a machine learning engineer, supporting data scientists' projects and career growth before its integration into the ML research team.
  • Prioritized dozens of data science projects and convinced stakeholders to terminate all but two.
Staff machine learning engineer Jan. 2020 - Mar. 2023
  • Participated in numerous ML projects, leading to improvements in various KPIs such as user engagement, cost reduction, and revenue growth.

BABLABS

CEO, co-founder Sep. 2015 - Dec. 2019
  • Developed and ran the university cafeteria menu information service, 밥대생 (ranked #1 in 2018, 500K MAU).
  • Responsible for Android app, web app, KakaoTalk chatbot, and crawling system.

Brew Glass

Director, co-founder Oct. 2014 - Aug. 2015
  • Developed the recommendation system and web app for the beer and whisky recommendation service, Glass.

Freelance

Android engineer Dec. 2011 - Sep. 2014
  • Developed 20+ Android applications.

Education

KAIST (Korea Advanced Institute of Science and Technology)
Master's degree, School of computing
Machine learning an intelligence laboratory (advisor: Eunho Yang)
KAIST (Korea Advanced Institute of Science and Technology)
Bachelor's degree, School of electrical and electronic engineering (double major in school of computing)
Algorithmic intelligence laboratory intern (advisor: Jinwoo Shin)
Mobile app development club: include

Publication

CUPID: A Real-Time Session-Based Reciprocal Recommendation System for a One-on-One Social Discovery Platform 2024
2+ author, International Conference on Data Mining (ICDM) Workshops, 2024
Learning with noisy labels by efficient transition matrix estimation to combat label miscorrection 2022
1st co-author, European Conference on Computer Vision (ECCV), 2022
Embedding Normalization: Significance Preserving Feature Normalization for CTR Prediction 2021
1st author, International Conference on Data Mining (ICDM) Workshops, Best Paper Awards, 2021
Efficient Click-Through Rate Prediction for Developing Countries via Tabular Learning 2021
1st author, International Conference on Learning Representations (ICLR) Workshops, Oral Presentations, 2021
Why Not to Use Zero Imputation? Correcting Sparsity Bias in Training Neural Networks 2020
1st author, International Conference on Learning Representations (ICLR), 2020