Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Data-centric MLOps(이정권)
Search
MLOpsKR
June 05, 2021
Programming
0
870
Data-centric MLOps(이정권)
MLOps KR(
https://www.facebook.com/groups/mlopskr)에서
주최한 1회 온라인 이벤트 발표 자료입니다
MLOpsKR
June 05, 2021
Tweet
Share
More Decks by MLOpsKR
See All by MLOpsKR
Ray: 대규모 ML인프라를 위한 분산 시스템 프레임워크(조상빈)
mlopskr
0
1.9k
JupyterFlow : 당신의 모델에 날개를 달아드립니다(유홍근)
mlopskr
0
1k
모델을 데이터셋에 맞게 대량을 찍어내는 방법(only 파이썬)(김태영)
mlopskr
0
760
KRSH: 선언형 Kubeflow, Terraform처럼 파이프라인 관리하기(김완수)
mlopskr
0
830
MLOps 춘추 전국 시대 정리(변성윤)
mlopskr
0
11k
Other Decks in Programming
See All in Programming
Webエディタライブラリ 「CodeMirror」から学ぶ Webアプリ開発のテクニック
ryosukeigarashi
0
250
Rubyのパフォーマンスプロファイリングの改善 / Enhancing performance profiling for Ruby
osyoyu
1
410
しくじり先生 Image Matching Challenge 2024 編
goosehaaan
0
810
Findy - エンジニア向け会社紹介 / Findy Letter for Engineers
findyinc
2
81k
CSC307 Lecture 10
javiergs
PRO
0
310
最古の関数型言語「Lisp」ことはじめ / lisp_in_kamiyama
uhooi
1
190
CSC307 Lecture 08
javiergs
PRO
0
330
Microservices rules (July 2024) : what good looks like
cer
PRO
0
1.6k
英語
s_shimotori
1
220
AWS CDKにおける「再利用性」を考える / aws-cdk-reusability
gotok365
6
1.3k
Terraformテスト入門
msato
0
550
小さな開発会社を作った理由
polidog
0
1.9k
Featured
See All Featured
A Philosophy of Restraint
colly
200
16k
Being A Developer After 40
akosma
72
580k
Stop Working from a Prison Cell
hatefulcrawdad
266
20k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
35
6.3k
From Idea to $5000 a Month in 5 Months
shpigford
377
46k
Raft: Consensus for Rubyists
vanstee
134
6.5k
Leading Effective Engineering Teams 2024
addyosmani
3
300
Practical Orchestrator
shlominoach
185
10k
What the flash - Photography Introduction
edds
65
11k
We Have a Design System, Now What?
morganepeng
46
7k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
662
120k
The Cost Of JavaScript in 2023
addyosmani
31
4.7k
Transcript
Data-centric MLOps : 데이터 중심 MLOps를 돕기 위한 작은 장치들
Superb AI 이정권
AI / ML = Model + Data
AI / ML = Model + Data Data centric?
Task Baseline: 70% accuracy Target Performance: 90% accuracy Should the
team improve the code or the data? : code(20%), data(80%) A Chat with Andrew on MLOps: From Model-centric to Data-centric AI
A Chat with Andrew on MLOps: From Model-centric to Data-centric
AI Improve AI → Improve the quality of the data: consistency error rate diversity coverage feedback frequency size ...
A Chat with Andrew on MLOps: From Model-centric to Data-centric
AI slide credit: A Chat with Andrew on MLOps: From Model-centric to Data-centric AI (https://www.youtube.com/watch?v=06-AZXmwHjo)
사실은, 늘 해오던 일 Project progress month 1 month 2
month 3 month 4 month 5 Code a model Build data Launch training job
사실은, 늘 해오던 일 Building the Software 2.0 Stack (Andrej
Karpathy, 2018)
Question: How many labeled images are needed to solve this
problem?
Answer: 100,000 images?
My Answer: I don’t know. Let’s start from 5,000 WHY?
여전히, 잘 모른다 → Data-centric MLOps Systematic & iterative way
to build Data for ML 단순히 지루한 작업을 자동화하는 과정이 아닌 ML 문제를 해결하기 위한 과정 저는 Superb AI라는 팀에서 이 문제를 풀고 있습니다.
<2달 <30명 <20,000 Images The Problem
The Meta Problem Design Data Spec Build Data Train a
model Deploy to service
Starting Point Labeling Tool Data Label
Reusable Data Spec { project_name: potato_detect_1 data_spec: good_potato: box: color:
red condition: ... bad_potato: box: } { project_name: potato_detect_2 data_spec: good_potato: polygon: color: red condition: ... bad_potato: box: }
Reusable Data Spec { project_name: potato_detect_13 data_spec: best_potato: polygon: direction:
options: ... good_potato: {} normal_potato: {} bad_potato: {} } Goal ≠ Task ALWAYS configured repeatedly name, color, type, conditions, options, property, ROI Info, ...
Support flexible pipeline 100 different problems, 100 different datasets, 100
different ways To support flexible pipeline Build Data Team Model WORKING SUBMITTED REVIEWED
Support flexible pipeline
Versioning Set 단위, 실험 당
ML Engineer를 위해 … ? Detailed Statistics & Report
Human in the loop ^ 2 Human in the loop
ML
Inside Human Labeling Data Human Labeling Service Model Data Labeling
Our Model ? Uncertain? Label-wise Confidence Overall Set Confidence User performance estimate Boost Labeling ... Human in the loop ^ 2
Keep labels consistent
Keep labels consistent
요약
Source data analysis, User analysis, Log, Task matching, etc 여전히
할일이 정말 많다. 마무리 SDK를 이용한 사용 예제!는 다음에 https://github.com/superb-AI-Suite/ Full-pipeline MLOps https://ai-infrastructure.org/