Jaehoon Ahn (안재훈)

PhD Candidate in Media Engineering · Sogang University · Seoul, South Korea

About

I am a PhD candidate at Sogang University advised by Prof. Moon-Ryul Jung, with coursework completed as of August 2025. My research sits at the intersection of machine learning and underserved domains: music information retrieval, low-resource NLP for ancient languages, motion synthesis, and content safety.

Alongside my graduate studies I spent three years shipping Roblox experiences at Dubit for clients including Viacom, Samsung, and SM Entertainment, among others, and most recently worked as an AI research engineer at Paralov. I hold a BS in Computer Science from Stony Brook University and am a US / South Korean dual citizen based in Seoul.


Publications

Appearance-Invariant Detection of Suggestive Motion via Laban Movement Descriptors on SMPL Skeletons
SIGGRAPH 2026 (Posters)
The SMC Blind Spot: Beat tracking failure mode analysis of hard music audio samples Under review
arXiv 2605.12287, 2026 · ISMIR 2026
BeatFCOS: Adapting object detection architecture for beat tracking in music audio
arXiv 2510.14391, 2025
ConTEXTure: Text-driven 3D texture generation via diffusion-based multi-view inverse rendering
arXiv 2407.10558, 2024

Selected Research

Extending LMs to Low-Resource Languages
Extending a multilingual language model to support low-resource languages with unseen scripts (Coptic, Syriac, Gothic, OCS Glagolitic, Samaritan). Results yield 8–106× perplexity reduction over conventional fine-tuning. Paper in preparation.
VBPM (Variational Bar Pointer Model)
Replaces the traditional Dynamic Bayesian Network in beat tracking pipelines with a fully differentiable variational model over structured latent variables, allowing for end-to-end training.
KineGuard
Skeleton-based content safety classifier for real-time unsafe motion detection. Built multi-GPU video processing pipeline with WHAM inference.

Experience

AI Research Engineer (Contract)
Paralov · Remote (Norway) · Dec 2025 – May 2026
Extended an AI-powered game generation pipeline for Roblox, integrating multimodal LLMs and image embeddings while engineering workarounds for closed-platform constraints.
Game Developer
Dubit · Remote (UK) · Jul 2021 – Jun 2024
Shipped Roblox experiences for clients including Viacom, Samsung, and SM Entertainment, among others. Tens of millions of unique player visits.
Software Engineering Intern
Roblox · San Mateo, US · Jun – Nov 2019
Implemented a redesigned mobile app landing page and login flow using Lua and React, with platform-specific fixes in Java, Objective-C, and C++.
Software Engineering Intern
DribbleUp · Brooklyn, US · May – Aug 2017
Developed features for a cross-platform React Native mobile app used by thousands of consumers.

Education

PhD Candidate in Media Engineering (ML/AI Focus)
Sogang University · Seoul, South Korea · Mar 2023 – Present
Advisor: Prof. Moon-Ryul Jung. Coursework completed August 2025.
MS in Media Engineering
Sogang University · Seoul, South Korea · Sep 2020 – Feb 2023
Thesis: "Adapting object detection techniques for beat tracking in musical audio."
BS in Computer Science
Stony Brook University · New York, US · Aug 2016 – May 2020

Skills

ML/DL: PyTorch, diffusion models, VAEs, transformers, object detection, LLM orchestration
Languages: Python, JavaScript/TypeScript, Lua/Luau, Java, C#, C/C++
Infrastructure: Multi-GPU training/inference, CUDA, Docker, Git, REST APIs
Spoken: Bilingual English/Korean; Mandarin Chinese (B1)