Skip to content
View lastdefiance20's full-sized avatar

Highlights

  • Pro

Block or report lastdefiance20

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Please don't include any personal information such as legal names or email addresses. Maximum 100 characters, markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
lastdefiance20/README.md

Hi, I'm Jaeyoon Jung

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.

🏒 Career

  • Misys Lab intern (2021-06-18 ~ 2022-08)
  • πŸ’Ό 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

πŸ“ Publications

πŸ™Œ EnCLAP: Combining Neural Audio Codec and Audio-Text Joint Embedding for Automated Audio Captioning

Accepted to ICASSP 2024
paper | code

πŸ† AI Challenge

πŸ₯‡ [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

πŸ₯‡ [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

πŸ₯ˆ [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)

πŸŽ–οΈ [4th Place] in Monthly Dacon Computer Vision Anomaly Detection

Task: detect the anomaly samples and classify it
code link

πŸŽ–οΈ [6th Place] in Monthly Dacon 3D MNIST Classification

Task: 3D MNIST Classification
repo link

πŸŽ–οΈ [7th Place] in 2022 SWUNIV AI Challenge

Task: develop OCR algorithm to recognize hangul text from the image
repo link | soongsil univ news

πŸŽ–οΈ [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 | Open In Colab

[Reach the final] in 2022 Military AI Competition

repo link

[Reach the final] in 2022 Naver AI RUSH

repo link

πŸ“’ AI Project

πŸ₯ˆ [2nd Prize] 2022 Soongsil univ AI Contest

'TryYours' the high resolution virtual try on using HR-VITON
repo link | Open In Colab

πŸŽ–οΈ [participation Prize] 2021 Soongsil univ AI Contest

'Jaeho' the AI speaker that has its name and facial expression
repo link

πŸ’» Algorithm Problem Solving

  • BOJ (Baekjoon Online Judge)

Solved.ac ν”„λ‘œν•„

Pinned Loading

  1. TryYours-Virtual-Try-On TryYours-Virtual-Try-On Public

    Virtual Try On site using HR-VITON. Anyone can try it using their own Image with clothes image.

    Python 102 39

  2. ML_Competition ML_Competition Public

    kaggle(캐글), dacon(데이콘) λ“± 데이터 κ²½μ§„λŒ€νšŒ μ°Έκ°€ μ½”λ“œ μ €μž₯μ†Œ

    Jupyter Notebook 3 6

  3. 2022-Samsung-AI-Challenge-3D-Metrology-1st-place-Solution 2022-Samsung-AI-Challenge-3D-Metrology-1st-place-Solution Public

    2022 Samsung AI Challenge (3D Metrology) 1st place Solution

    Jupyter Notebook 4 3

  4. 2023-LG-DISPLAY-Quality-Classification-1st-place-Solution 2023-LG-DISPLAY-Quality-Classification-1st-place-Solution Public

    2023 LG DISPLAY Product Quality Classification 1st place Solution

    Jupyter Notebook 3 2

  5. Bookathon2022_Team_Booker2 Bookathon2022_Team_Booker2 Public

    Sungkyunkwan Univ's 4th Bookathon. Write essay with AI

    Python 7

  6. worv-ai/canvas worv-ai/canvas Public

    CANVAS: Commonsense-Aware Navigation System for Intuitive Human-Robot Interaction

    8 1