As an undergraduate student majoring in Artificial Intelligence, I am interested in diverse research areas, including computer vision, natural language processing, and audio signal processing.
My current research interests are several. Large language models that has capability of multimodal inputs and outputs (video, image, text, audio, action), adapting instruction-following capabilities of English-based large language models to mainly Korean and other diverse language, and dehallucinative large language model.
- Misys Lab intern (2021-06-18 ~ 2022-08)
- πΌ Maum AI Inc (Mindslab) AI Scientist (2023-04-26 ~ )
- Improved the quality of RAG by enhancing the retriever. By training it with a high-quality dataset and selecting the best models, the retriever achieved maximum 68% score improvement(Recall@K) compared to its previous performance.
- Optimized STF(Speech-To-Face) model by using ONNX and TensorRT, reducing the 1-second video generation time from 1.6 seconds to 0.7 seconds. This enhancement enables real-time streaming in real-world applications.
- Worked on transferring instruction-following capabilities from English to Korean in open-source large language models. This work aims to facilitate the development of high-quality Korean language models at a low cost.
- Experienced with training large language models(up to 70B), using deepspeed for multi-node training. While each node comprises 8 NVIDIA H100 80GB GPUs, four DGX H100 systems interconnected with NVLink are used for train.
- https://maum-ai.github.io
π EnCLAP: Combining Neural Audio Codec and Audio-Text Joint Embedding for Automated Audio Captioning
π₯ [1st Place] in 2022 Samsung AI Challenge (3D Metrology)
Task: make an AI that produce depth map from SEM image
repo link | soongsil univ news
π₯ [1st Place] in 2023 LG DISPLAY Product Quality Classification
Task: classify the product quality using tabular data from the LG display factory
repo link | hankyung news
π₯ [2nd Place] in 2022 LG INNOTEK Radar Performance Prediction
Task: predict radar performance using tabular data from the LG innotek factory
repo link | youtube link (interview)
ποΈ [4th Place] in Monthly Dacon Computer Vision Anomaly Detection
Task: detect the anomaly samples and classify it
code link
ποΈ [6th Place] in Monthly Dacon 3D MNIST Classification
Task: 3D MNIST Classification
repo link
ποΈ [7th Place] in 2022 SWUNIV AI Challenge
Task: develop OCR algorithm to recognize hangul text from the image
repo link | soongsil univ news
ποΈ [7th Place] in 2022 Dankook Univ AI Challenge
Task: predict bike sharing demand using tabular data
repo link
[Reach the final] in 2023 4th Sungkyunkwan Univ Bookathon
Task: write essay with AI (GPT3)
repo link |
[Reach the final] in 2022 Military AI Competition
[Reach the final] in 2022 Naver AI RUSH
π₯ [2nd Prize] 2022 Soongsil univ AI Contest
'TryYours' the high resolution virtual try on using HR-VITON
repo link |
ποΈ [participation Prize] 2021 Soongsil univ AI Contest
'Jaeho' the AI speaker that has its name and facial expression
repo link