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

[2022/09/01]Machine Learning 輪講 #158

Open hakubishin3 opened 1 year ago

hakubishin3 commented 1 year 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. https://github.com/wantedly/machine-learning-round-table/issues/156

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 1 year ago

Matching Theory-based Recommender Systems in Online Dating

日本語紹介記事: https://www.cyberagent.co.jp/news/detail/id=27897

nogawanogawa commented 1 year ago

ItemSage: Learning Product Embeddings for Shopping Recommendations at Pinterest

論文URL

https://dl.acm.org/doi/10.1145/3534678.3539170

著者

Paul Baltescu, Haoyu Chen, Nikil Pancha, Andrew Zhai, Jure Leskovec, Charles Rosenberg

会議

KDD '22

背景

Pinterestでは、画像のシェア・閲覧に加えて、商品の購入が可能になっている。 Pinterestの主要コンテンツであるピンとは異なり、商品には様々な角度や文脈から撮影された複数の画像で構成され、タイトル、説明、色、柄などのメタデータが提供されており、データそのものの性質がPinと商品では異なる。

また、商品購入はクリックやブックマークなどに加えて、カートに追加・購入など、Pinとは異なるタイプのアクションに最適化した推薦が必要になる。

目的

アプローチ

image

image

memo

https://github.com/nogawanogawa/paper_memo/issues/54

zerebom commented 1 year ago

OptFormer: Towards Universal Hyperparameter Optimization with Transformers

blog URL: https://ai.googleblog.com/2022/08/optformer-towards-universal.html image Googleの論文。ハイパラチューニングにTransformerを使った手法。meta dataを食わせ、パラメータを生成させ、精度を予測するタスクを解かせる。

全てのタスクが全く同じハイパラメータ探索空間を使用しなくて良い、数値だけでなくカテゴリ変数(ex. どのoptimzerを使うか) にも対応できる、といったメリットがあるみたい