Career Profile
I am a ML research engineer with a PhD in Computational Linguistics, turning advances in language and agent models into practical products. At Samsung Electronics, I develop machine learning models for CTR, user response prediction, and conversion-based advertising on Samsung TV Plus. I previously contributed to the global launch of a TV agent system using Google Gemini on Google Cloud. My research experience includes internships at Meta GenAI and Amazon Alexa AI.
Experience
Samsung Ads: Improving conversion-based advertising with modern AI techniques Feb 2026 - Present
- Develop advertising machine learning models for the Samsung TV Plus platform using modern AI techniques, including Transformer-based models.
- Build models for click-through rate (CTR) and user response prediction.
- Work on connected TV (CTV)-to-mobile conversion advertising.
Tech stack: AWS, TensorFlow, PyTorch, Snowflake, Grafana
Integrating LLM-based agent systems into Samsung TV technology Feb 2025 - Jan 2026
- Contributed to the global launch of a TV agent system in July 2025, deploying Google Gemini on Google Cloud Platform (GCP).
Tech stack: Google Gemini, Google Cloud, Vertex AI, Python, LangChain
- Researched pretraining of inference-efficient audio language models using alternative Transformer architectures and quantization.
Tech stack: PyTorch, Hugging Face, Fairseq, DeepSpeed, Python
- Studied the interpretability of neural models using attention and grammatical information for natural language understanding and skill routing.
Tech stack: AWS, Hugging Face, Python
- Project: Multi-Document Summarization with External Knowledge; Tobias Falke, Prof. Iryna Gurevych [slides]
- Project: iASIS Knowledge Graph; Prof. Soren Auer
- Delivered machine learning systems for government institutes and critical public infrastructure.
- Designed and evaluated systems using audio, web-page, and network traffic data.
- Work resulted in three US patents and two first-author research papers.
Education
- Thesis: Neural Centering-based Coherence Modeling
- Advisor: Prof. Michael Strube
- Thesis: Spoiler Detection in TV Program Tweets (published at ICWSM 2013)
- Best Graduation Project: Detecting Counter Attacks from a Soccer Video
Selected Publications
Entity Tracking in Small Language Models: An Attention-Based Study of Parameter-Efficient Fine-Tuning
Sungho Jeon, Michael Strube
CODI @ EMNLP, 2025.
An evaluation framework for entity tracking that measures attention flow between entity and non-entity tokens in small language models, before and after parameter-efficient fine-tuning.
Attention or Convolution: Transformer Encoders in Audio Language Models for Inference Efficiency
Sungho Jeon, Ching-Feng Yeh, Hakan Inan, Wei-Ning Hsu, Rashi Rungta, Yashar Mehdad, Daniel Bikel
SASB @ ICASSP, 2024.
Efficient Transformer encoders for pretrained audio language models, including 1-bit weight quantization.
Countering the Influence of Essay Length in Neural Essay Scoring
Centering-based Neural Coherence Modeling with Hierarchical Discourse Segments
Incremental Neural Lexical Coherence Modeling
Skills & Service
Tech stack: PyTorch, TensorFlow, Hugging Face, Python, LangChain, Google Cloud, Vertex AI, AWS, Snowflake, Grafana
Peer reviewer: EMNLP and ACL since 2021; COLING and CODI since 2020
Service: Publicity & Infrastructure Chair, EACL 2024