wantedly / machine-learning-round-table

Gather around the table, and have a discussion to catch up the latest trend of machine learning 🤖
https://www.wantedly.com/projects/391912
297 stars 2 forks source link

[2021/07/08]Machine Learning 輪講 #107

Open hakubishin3 opened 3 years ago

hakubishin3 commented 3 years ago

Why

Machine Learning 輪講は最新の技術や論文を追うことで、エンジニアが「技術で解決できること」のレベルをあげていくことを目的にした会です。

prev. #105

What

話したいことがある人はここにコメントしましょう! 面白いものを見つけた時点でとりあえず話すという宣言だけでもしましょう!

hakubishin3 commented 3 years ago

Using AntiPatterns to avoid MLOps Mistakes

https://arxiv.org/abs/2107.00079

MLOpsにおけるアンチパターンを紹介している. (思ったよりもモデリングの比重が大きかった)

## Innovating Faster on Personalization Algorithms at Netflix Using Interleaving https://netflixtechblog.com/interleaving-in-online-experiments-at-netflix-a04ee392ec55 ## Comparing the sensitivity of information retrieval metrics https://dl.acm.org/doi/10.1145/1835449.1835560 ## A probabilistic method for inferring preferences from clicks https://dl.acm.org/doi/10.1145/2063576.2063618
zerebom commented 2 years ago

CUPED

http://robotics.stanford.edu/people/ronnyk.link/2013-02CUPEDImprovingSensitivityOfControlledExperiments.pdf A/Bテスト時に、目的変数の分散を小さくすることで検出力を上げる手法。 A/Bテスト間でバイアスのない共変量(過去の実験の目的変数など)用いて目的変数を修正する。 目的変数の分散を、共変量の回帰直線との二乗誤差として計算することで小さくする。

image 引用元: https://booking.ai/how-booking-com-increases-the-power-of-online-experiments-with-cuped-995d186fff1d

過去の勉強会で紹介された下記スライドでも言及されている。 https://speakerdeck.com/shyaginuma/btesutoniokeruvariance-reduction

その他解説記事。

nogawanogawa commented 2 years ago

Efficient Deep Learning: A Survey on Making Deep Learning Models Smaller, Faster, and Better

https://arxiv.org/abs/2106.08962

この辺を工夫しつつ、モデルの精度とのトレードオフを考慮することが良さそう

Be Careful When Interpreting Predictive Models in Search of Causal Insights

https://towardsdatascience.com/be-careful-when-interpreting-predictive-models-in-search-of-causal-insights-e68626e664b6

notebook: https://shap.readthedocs.io/en/latest/example_notebooks/overviews/Be%20careful%20when%20interpreting%20predictive%20models%20in%20search%20of%20causal%C2%A0insights.html

個人的に、解釈性についてあまり詳しくなかったので、使い方含め今度なぞってみたい