Seungyoun Shin(신승윤)

I am a Machine Learning Engineer at ChannelTalk, where I'm actively involved in developing and optimizing AI solutions.

My current focus includes:

Researching advanced function calling techniques in Large Language Models (LLMs), utilizing SFT and Preference Learning
Exploring synthetic data collection methods and seamless training approaches for enhanced model performance

I hold a Master's degree in Artificial Intelligence from Korea University, where I studied under professor Sungjoon Choi.

For professional inquiries or collaborations, please contact me at robin@channel.io.

Research Interest:

Large Language Models (LLMs), Synthetic Data Generation, Multimodal, Efficient Fine-tuning Strategies

Work Experience:

2024.06 - Present : Channel Talk, Machine Learning Engineer
2023.08 - 2024.01: NAVER Cloud, Creativity Lab Intern
2019.06 - 2019.09: mAyI, Software Engineer Intern

Education

M.S. in Artificial Intelligence, Korea University (Sep 2022 - Present). Advisor: Prof. Sungjoon Choi.
B.Sc. in Computer Science, Dongguk University (Mar 2018 - Aug 2022).

publications [full list]

2024

RAL

Towards Embedding Dynamic Personas in Interactive Robots: Masquerading Animated Social Kinematics (MASK)

Jeongeun Park, Taemoon Jeong, Hyeonseong Kim and 7 more authors

In

Blog

2023

NeurIPSWorkshop

Past as a Guide: Leveraging Retrospective Learning for Python Code Completion

Seungyoun Shin, Seunggyu Chang, and Sungjoon Choi

In NeurIPS 2023 Workshop on Instruction Tuning and Instruction Following

Abstract PDF Blog Code

This work presents Past as a Guide (PaG), a simple approach for Large Language Models (LLMs) to improve the coding capabilities by integrating the past history with interactive and iterative code refinements. To be specific, inspired by human cognitive processes, the proposed method enables LLMs to utilize previous programming and debugging experiences to enhance the Python code completion tasks. The framework facilitates LLMs to iteratively refine the Python code based on previous execution and debugging results and optimize learning and reasoning capabilities. The proposed methodology achieved a 92% pass@1 on HumanEval, demonstrating the potential to advance the field by leveraging retrospection from past experiences and interactive and iterative refinement processes without external correctness indicators.
ACPR2023

Towards Explainable Computer Vision Methods via Uncertainty Activation Map

Seungyoun Shin, Wonho Bae, Junhyug Noh and 1 more author

In ACPR 2023:Pattern Recognition

Abstract

This paper focuses on the problem of highlighting the input image regions that result in increasing predictive uncertainty. In particular, we focus on two types of uncertainty, epistemic and aleatoric, and present an uncertainty activation mapping method that can incorporate both types of uncertainty. To this end, we first utilize a mixture-of-experts model combined with class-activation mapping (CAM). The proposed method is extensively evaluated in two different scenarios: multi-label and artificial noise injection scenarios, where we show that our proposed method can effectively capture uncertain regions
ACPR2023

Robust Detection for Autonomous Elevator Boarding Using a Mobile Manipulator

Seungyoun Shin, Joon Hyung Lee, Junhyug Noh and 1 more author

In ACPR 2023:Pattern Recognition

Abstract Blog

Indoor robots are becoming increasingly prevalent across a range of sectors, but the challenge of navigating multi-level structures through elevators remains largely uncharted. For a robot to operate successfully, it’s pivotal to have an accurate perception of elevator states. This paper presents a robust robotic system, tailored to interact adeptly with elevators by discerning their status, actuating buttons, and boarding seamlessly. Given the inherent issues of class imbalance and limited data, we utilize the YOLOv7 model and adopt specific strategies to counteract the potential decline in object detection performance. Our method effectively confronts the class imbalance and label dependency observed in real-world datasets, Our method effectively confronts the class imbalance and label dependency observed in real-world datasets, offering a promising approach to improve indoor robotic navigation systems.

2022

arXiv

Learning Joint Representation of Human Motion and Language

Jihoon Kim, Youngjae Yu, Seungyoun Shin and 2 more authors

arXiv:2210.15187

PDF
PR

Conditional motion in-betweening

Jihoon Kim, Taehyun Byun, Seungyoun Shin and 2 more authors

Pattern Recognition

PDF Blog
PR

Elucidating robust learning with uncertainty-aware corruption pattern estimation

Jeongeun Park, Seungyoun Shin, Sangheum Hwang and 1 more author

Pattern Recognition

PDF