Skip to content
View woongjoonchoi's full-sized avatar
๐ŸŽฏ
Focusing
๐ŸŽฏ
Focusing

Block or report woongjoonchoi

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.

Maximum 250 characters. Please donโ€™t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this userโ€™s behavior. Learn more about reporting abuse.

Report abuse
woongjoonchoi/README.md

Hi there ๐Ÿ‘‹

I'm Research scientist who specializes in system-algo co-design aware hardware.

TechStack

OpenSource Contribution

1 . Pytorch Tutorial ์˜คํƒ€ ์ˆ˜์ • pytorch/tutorials#1845
2 . Pytorch Md5 Checksum issue find pytorch/vision#8220
3 . Pytorch Tutorial Contribution pytorch/tutorials#2971

Carrer

์ž์„ธํžˆ ๋ณด๊ธฐ

Researcher

KIST

2024.Sep -

  • ์ปดํ“จํ„ฐ ๋น„์ „ ๊ฒฝ๋Ÿ‰ํ™” ์—ฐ๊ตฌ
  • MLSystem(ex.kernel implementation,Distribute training,Distributed Inference)์—ฐ๊ตฌ

Naver boostcamp AI Tech

2021.Aug - 2022.Jan

  • 3๊ฐœ์˜ Team์„ ์ ๊ทน์ ์œผ๋กœ ์ด๋Œ๋ฉฐ , ๋ฌธ์„œ ๊ด€๋ฆฌ ๋ฐ GIt์„ ์‚ฌ์šฉํ•œ ํ”„๋กœ์ ํŠธ ๊ด€๋ฆฌ. ์ด๋ฅผ ํ†ตํ•ด ์ฃผ๋„์ ์œผ๋กœ ์›๊ฒฉํ˜‘์—…์„ ์ด๋Œ์—ˆ์Œ.
  • ๋‚ด๋ถ€ Competetion์—์„œ Team์„ ๋ฆฌ๋“œํ•˜์—ฌ , ๊ฐ๊ฐ 3๋“ฑ , 6๋“ฑ์˜ ์„ฑ๊ณผ๋ฅผ ๋‹ฌ์„ฑํ•จ
  • ๊ธฐ์ดˆ Coruse์—์„œ Math,Python,Data Visualization , DeepLearning Architecture ๊ด€๋ จ ์ง€์‹์„ ์Šต๋“ํ•จ.
  • NLP Course Track์„ ์„ ํƒํ•˜์—ฌ LLM,ODQA,Relation Extraction ๋“ฑ์˜ task๋ฅผ ํ•™์Šตํ•˜์˜€์Œ.
  • ๋‚ด๋ถ€ ์ปค๋ฎค๋‹ˆํ‹ฐ์—์„œ ๋งค์ผ ๊ธ€์“ฐ๊ธฐ ํ™œ๋™ ๋ฐ Q&A ํ™œ๋™์„ ์ ๊ทน์ ์œผ๋กœ ์ž„ํ•˜์—ฌ BoostTech๋ฅผ ๋น›๋‚ธ ์บ ํผ์— ์„ ์ •๋จ

Google MachineLearning Bootcamp

2021.Aug - 2021.DEC

  • Coursera์˜ DeepLearning Specialization Course๋ฅผ ์ˆ˜๊ฐ•ํ•˜์—ฌ ๋”ฅ๋Ÿฌ๋‹ task์— ๋Œ€ํ•ด ์ด๋ก ์ ์ธ ์ง€์‹์„ ๊ฐ•ํ™”ํ•จ
  • Google Cloud์˜ Professional Data Engineer ์ž๊ฒฉ์ฆ์„ ์ทจ๋“ํ•จ.

Personal Project

์ž์„ธํžˆ ๋ณด๊ธฐ

DeepLearning Paper Reproducing (2023.11~)

  • DeepLearning Paper์˜ ๋ชจ๋“  configuration์„ ๋ณต์ œํ•˜์—ฌ ๋…ผ๋ฌธ์˜ ์„ฑ๋Šฅ์„ ์žฌํ˜„ํ•˜๊ณ ์ž ํ•จ.
  • 140M ์˜ parameter๋ฅผ ๊ฐ€์ง€๋Š” VGG model์„ 140GB ์ธ ImageNet ๋ฐ์ดํ„ฐ์— ๋Œ€ํ•˜์—ฌ scratch๋ถ€ํ„ฐ ํ›ˆ๋ จํ•˜์—ฌ ๋…ผ๋ฌธ์˜ ์„ฑ๋Šฅ๊ณผ 3% ์˜ ์˜ค์ฐจ๋ฒ”์œ„ ๋‚ด์— ์ˆ˜๋ ด์‹œํ‚ด.

AI Paperboy (2021.10~2021.12)

  • 4๋ช…์˜ ํŒ€์„ ๋ฆฌ๋“œํ•˜๊ณ  , ์Šฌ๋ž™์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ์ธ ์—…๋ฌด ๊ด€๋ฆฌ, ์คŒ์„ ์‚ฌ์šฉํ•˜์—ฌ ๋ฏธํŒ…์„ ์ง„ํ–‰ , ๊ตฌ๊ธ€ ๋“œ๋ผ์ด๋ธŒ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ฌธ์„œ ๊ด€๋ฆฌ ๋ฐ GIt์„ ์‚ฌ์šฉํ•œ ํ”„๋กœ์ ํŠธ ๊ด€๋ฆฌ. ์ด๋ฅผ ํ†ตํ•ด ์ฃผ๋„์ ์œผ๋กœ ์›๊ฒฉํ˜‘์—…์„ ์ด๋Œ์—ˆ์Œ. ์œ ์ €๊ฐ€ ๋‰ด์Šค ๊ฒ€์ƒ‰ํ›„ ๊ด€๋ จ ๋‰ด์Šค๋ฅผ ๊ฒ€์ƒ‰ํ•˜๋Š” ๊ณผ์ •์„ ์ค„์ด๊ธฐ ์œ„ํ•ด์„œ ๊ด€๋ จ ๋‰ด์Šค ์Šค๋‹ˆํŽซ์„ ์ €์žฅํ•˜๋Š” LLM Aplication์„ ๊ฐœ๋ฐœํ•ด์„œ ๊ด€๋ จ ๋‰ด์Šค ๊ฒ€์ƒ‰์˜ 4๋‹จ๊ณ„ ๊ณผ์ •์„ 1๋‹จ๊ณ„๋กœ ์ค„์—ฌ์„œ ๊ฒ€์ƒ‰์‹œ๊ฐ„์„ 40% ๊ฐ์†Œ.
  • 84๋งŒ๊ฐœ์˜ ๋‰ด์Šค ๋ฐ์ดํ„ฐ๋ฅผ ์ˆ˜์ง‘ํ•˜์—ฌ ๊ฐœ์ธ์ •๋ณด,์ €์ž‘๊ถŒ,ํŠน์ˆ˜๋ฌธ์ž ๋“ฑ์„ ์ •๊ทœํ‘œํ˜„์‹์„ ์‚ฌ์šฉํ•˜์—ฌ ์ œ๊ฑฐ ํ•˜๊ณ  ๋งž์ถค๋ฒ•์„ ๊ต์ •ํ•˜๋Š” ์ „์ฒ˜๋ฆฌ ์ง„ํ–‰
  • Huggingface์—์„œ ์ œ๊ณตํ•˜๋Š” klue/roberta-large model์„ Pytorch์—์„œ ์ˆ˜์ง‘ํ•œ ๋‰ด์Šค ๋ฐ์ดํ„ฐ๋กœ fine-tuning ํ•ด์„œ ODQA model์„ ๊ตฌํ˜„ .Weight & Bias ํ”Œ๋žซํผ์—์„œ Bayesian Search๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ Hyperparameter search๋ฅผ ์ˆ˜ํ–‰. baseline ๋Œ€๋น„ 30%์˜ ์„ฑ๋Šฅ ํ–ฅ์ƒ
  • User flow์™€ Data Flow๋ฅผ ์ž‘์„ฑํ•˜์—ฌ specification์„ ๋งŒ๋“ค๊ณ  , fastapi๋ฅผ ํ†ตํ•ด์„œ ๋ชจ๋ธ์„ ์›นAPI๋กœ ๋งŒ๋“ ํ›„ GCP์˜ server์— ๋ฐฐํฌํ•จ
  • Github ๋งํฌ :๋งํฌ Youtube ๋งํฌ:๋งํฌ

Relation Extraction (2021.09~2021.10)

  • 5๋ช…์˜ ํŒ€์„ ๋ฆฌ๋“œํ•˜๊ณ  , ์คŒ์„ ํ†ตํ•ด ์›๊ฒฉ์œผ๋กœ ํšŒ์˜ ์ง„ํ–‰ ๋ฐ weight&bias ์—์„œ ํŒ€์˜ hyperparameter search ๋ฐ model evaluation ๊ฒฐ๊ณผ๋ฅผ ๊ด€๋ฆฌํ•˜์—ฌ ์ฃผ๋„์ ์œผ๋กœ ์›๊ฒฉํ˜‘์—…์„ ์ด๋Œ์—‡์Œ.
  • Competition์—์„œ ,๋ฌธ์žฅ์—์„œ 2๊ฐœ์˜ entity๊ฐ„์˜ ๊ด€๊ณ„๋ฅผ ๋ถ„๋ฅ˜ํ•˜๋Š” Model์„ Pytorch์—์„œ klue/Roberta-large-Model์„ Fine-tuningํ•˜์—ฌ ๊ฐœ๋ฐœํ•จ. Weight & Bias ํ”Œ๋žซํผ์—์„œ Bayesian Search๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ + Hyperparameter search๋ฅผ ์ˆ˜ํ–‰. 19๊ฐœ์˜ ์ฐธ๊ฐ€์กฐ ์ค‘ ์ตœ์ข… 6๋“ฑ์„ํ•จ. baseline-model์˜ error๋ฅผ 50% ์ž„ .
  • Github ๋งํฌ ppt ๋งํฌ

Mask Classification (2021.08~2021.08)

  • 7๋ช…์„ ๋ฆฌ๋“œํ•˜๊ณ  , ํŒ€ ์ „์ฒด์˜ ์ฝ”๋“œ ๋ฆฌ๋ทฐ๋ฅผ ๋‹ด๋‹นํ•˜๊ณ  GIt์„ ํ†ตํ•˜์—ฌ ํ”„๋กœ์ ํŠธ ๊ด€๋ฆฌ ๋ฐ ์คŒ์„ ํ†ตํ•˜์—ฌ ์›๊ฒฉํšŒ์˜ ์ง„ํ–‰์„ ํ•จ. ์ฃผ๋„์ ์œผ๋กœ ํŒ€ ์ „์ฒด์˜ ์ฝ”๋“œ ์•„ํ‚คํ…์ณ๋ฅผ ํ†ต์ผํ•˜๊ณ  ์›๊ฒฉํ˜‘์—…์„ ์ด๋Œ์—ˆ์Œ.
  • Competetion์—์„œ ๋‚˜์ด,์„ฑ๋ณ„,๋งˆ์Šคํฌ ์ฐฉ์šฉ์„ ํ™•์ธํ•˜๋Š”Image Classification ๋ชจ๋ธ์„ Pytorch์—์„œ ๊ตฌํ˜„ํ•˜์—ฌ , 39๊ฐœ์˜ ์กฐ์ค‘ 8๋“ฑ์ด๋ผ๋Š” ์„ฑ๊ณผ๋ฅผ ์–ป์—ˆ์Šต๋‹ˆ๋‹ค. Python์˜ sequence type์ด ์‚ฌ์šฉ๋œ ์ฝ”๋“œ๋ฅผ generator type๋กœ ์ˆ˜์ •ํ•˜์—ฌ ๊ธฐ์กด ์ฝ”๋“œ์˜ Memory ์‚ฌ์šฉ๋Ÿ‰์„ 1/3์œผ๋กœ ์ค„์—ฌ์„œ ์ตœ์ ํ™”๋ฅผ ํ–ˆ์Šต๋‹ˆ๋‹ค.
  • Github ๋งํฌ

์•…ํ”Œํƒ์ง€ ์‹œ์Šคํ…œ (2020.03~2020.11)

  • ๋ฌด๋ถ„๋ณ„ํ•œ ์•…์„ฑ ๋Œ“๊ธ€์— ๊ณ ํ†ต๋ฐ›๋Š” ์‚ฌ๋žŒ๋“ค์„ ๋„์™€์ฃผ๋Š” LLM application์„ ๊ฐœ๋ฐœ. 100๋งŒ๊ฐœ์˜ ๋Œ“๊ธ€ ๋ฐ์ดํ„ฐ๋ฅผ ํฌ๋กค๋งํ•˜์—ฌ ์ •๊ทœ์‹์œผ๋กœ ์ „์ฒ˜๋ฆฌํ•˜๊ณ  Bert model์„ Tensorflow์—์„œ large scale trainingํ•˜์—ฌ Sentiment Classifier ๋ชจ๋ธ์„ ๊ฐœ๋ฐœ. Pretrained Huggingface model ๋Œ€๋น„ ์„ฑ๋Šฅ์ด 30% ์ฆ๊ฐ€. ๋‹ด๋‹น๊ต์ˆ˜๋‹˜์ด ๋‹ด๋‹นํ•˜๋Š” 3๊ฐœ์˜ ํŒ€์ค‘์—์„œ 1๋“ฑ์„ ํ•ด์„œ ํ•™๊ณผ ์ตœ์ข…๋ฐœํ‘œํšŒ์—์„œ ๋ฐœํ‘œ.
  • ์กธ์—…๋…ผ๋ฌธ๋งํฌ

Face Recognition & Verfication(2020.03~2020.07)

  • Tensorflow์˜ API๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ Object Detection Application ์„ ์ž‘์„ฑํ•˜์—ฌ ์ •ํ™•๋„ 90 %๋ฅผ ๋‹ฌ์„ฑ
  • Binary Classifiaction์œผ๋กœ Face Verification Model์„ scratch๋ถ€ํ„ฐ ์ž‘์„ฑํ•˜์˜€์œผ๋‚˜ ์‹คํŒจ. ์—ฌ๊ธฐ์„œ, Pretrained model์˜ ์ค‘์š”์„ฑ์„ ๊นจ๋‹ฌ์Œ
  • Github๋งํฌ:(https://github.com/woongjoonchoi/final-project-level3-nlp-19) ,๋ธ”๋กœ๊ทธ๋งํฌ:(https://woongjun-warehouse.tistory.com/25)

OCW and Mooc

๊ฐœ์ธ์ ์œผ๋กœ ๊ณต๋ถ€ํ•œ OCW,Mooc and STEM books.

์ž์„ธํžˆ ๋ณด๊ธฐ

DeepLearning Specialization :

certificate(link )

Assignment(๋ฒ„ํŠผํด๋ฆญ)
  1. Optimization Assignment from scratch - Korean

    Optimization Assignment from scratch - English

  2. Convolution Assignment from scratch - Korean

    Convolution Assignment from scratch - English

  3. FeedForward Math derivation - korean

    FeedForward Math derivation - english

NoteTaking(๋ฒ„ํŠผํด๋ฆญ)
  1. Structuring your machine learning projects
    Link
  2. Optimization,HyperParameter Tuning Link
  3. Convolution Neural Network Link
  4. Sequence Model

Pytorch

NoteTaking(๋ฒ„ํŠผํด๋ฆญ)

data api

Link Kor
Link Eng

MIt 6.006(Introduction to Algorithm):

NoteTaking(๋ฒ„ํŠผํด๋ฆญ)

lec 09 DFS and Topological Order

Link_Kor
Link_Eng

Assignment(๋ฒ„ํŠผํด๋ฆญ)

Problem5

GithubLInk

Berkely CS 162 :

Khan Academy Statistics:

3b1b Linear Algebra

์ž์„ธํžˆ๋ณด๊ธฐ

lec01~05(Vector Space,Linear Transformation)

Kor
Eng

Mit Linear Algebra 18.06

3b1b Calculus

Python

Python self-studying by STEM Book and Cpython github source.

์ž์„ธํžˆ๋ณด๊ธฐ

Learning Python

STEM Books about Python Beginner ~ Intermediate

Note Taking

All Post

Kor Link
Eng Link

Chp 4 Built-in Objects

Kor Link
Eng Link

Chp 5 Numeric

Kor Link
Eng Link

Chp 13 Loop

Kor Link
Eng Link

Chp 17 Scope

Kor Link
Eng Link

Python Performance in terms of Python internal implementation

Python Peformance๋ฅผ ๋‚ด๋ถ€ ๊ตฌํ˜„ ๊ด€์ ์—์„œ ๋ฐ”๋ผ๋ด…๋‹ˆ๋‹ค.

Note Taking

Total Link about python internal

InternalAll_Kor

InternalAll_Eng

Link About Python Integer Internal

Integer_internal_kor

Integer_internal_English

Link About Python String operation and method Internal

String method internal kor

String method internal Eng

Git-SCm

STEM book about git in terms of Distributed Version control system

์ž์„ธํžˆ๋ณด๊ธฐ

Note-taking

Pinned Loading

  1. woongjoonchoi.github.io woongjoonchoi.github.io Public

    HTML 1

  2. DeepLearningPaper-Reproducing DeepLearningPaper-Reproducing Public

    Jupyter Notebook

  3. OCW-and-MOCC OCW-and-MOCC Public

    Jupyter Notebook

  4. CodingTest CodingTest Public

    Python 1

  5. pytorch/tutorials pytorch/tutorials Public

    PyTorch tutorials.

    Python 9.1k 4.4k