권재명 박사의 카이스트 세미나

Leading AI + Data Science Team
Sep 2018
Jamie Kwon(https://dataninja.me)

AI > ML > DL

www.matroid.com/scaledml/2017/jeff/pdf
데이터 사이즈가 클수록 정확도는 어떻게 되는가?

https://www.cbinsight.com
www.hanalyze.com/2018/

AI + DS Project Examples
Ex1. Top n pages

https://www.evanmiller.org/how-not-to-sort-by-average-ratin…

Wilson Confidence interval

Ex2. Running A/B Test
https://www.evanmiller.org/sequential-ab-testing.html
http://pages.optimizely.com/rs/op...

Ad Click Prediction : a View from the Trenches <- paper

https://ai.google/reserach/pubs/pub41159
Algorithm1 : Per - Coordinate FTRL-Proximal with L1 and L2 regularization

Low-Rank Matrix Factorization

classification, regression, clustering, dimensionality reduction
Complication

Configuration, Data Collection, Data Verification, Machine Resource Management, ...

http://www.longtail.com/about.html

Common AI + DS tool and workflow
python 3.6 ecosystem
Scikit-learn +

R ecosystem
R Studio + tidyverse(+Jupyter)
www.stroybench.org

Jupyter Ecosystem
https://medium.com/netflix...

Tensorflow for DL

AI + DS Team
Profile of an analyst
-Basic Stats
-SQL
-Excel(Tableau)
-Attention to details
-Good Communication

Builing and Running AI + DS Team
1. Version control AI + DS codes
2. blogd.codinghorror.com/when-understanding-means-rewriting
Strive for readable codes
3. Strive for reproducibility
How R helps Aribnb make the most of its idea
5. Iterate fast
6. Produce high-quality power point
7. Open by default