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

[2023/07/19]Machine Learning 輪講 #204

Open hakubishin3 opened 12 months ago

hakubishin3 commented 12 months 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. #202

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!

tmiya commented 11 months ago

Countering Popularity Bias by Regularizing Score Differences, Recsys 2022 の話をしようと思います。

発表資料 -> https://github.com/tmiya/tmiya.github.io/blob/main/paper_reading/20230719.pdf

nogawanogawa commented 11 months ago

Diffusion Recommender Model

論文URL

https://arxiv.org/abs/2304.04971

著者

Wenjie Wang, Yiyan Xu, Fuli Feng, Xinyu Lin, Xiangnan He, Tat-Seng Chua

会議

SIGIR'23

背景

生成的推薦モデルは、未観測のインタラクションアイテムに関する確率を推論するための生成過程を学習している。 生成的推薦モデルはこれまでにも研究が進められてきた。

VAEの方が優れた結果を示しているものの、

の観点で問題が残っている。

一方、生成モデルとして、近年では拡散モデルが登場している。 順方向のプロセスで徐々にノイズを加えていき、逆方向プロセスで入力を再現するというプロセスで学習を行う。

推薦は、欠損した事前分布から事後分布を予測する逆方向プロセスに似ているため、ここに応用できる可能性がある。

目的

アプローチ

image

memo

https://github.com/nogawanogawa/paper_memo/issues/92

zerebom commented 11 months ago

VizCommender: Computing Text-Based Similarity in Visualization Repositories for Content-Based Recommendations

image
takashi-miyamoto-naviplus commented 11 months ago

Countering Popularity Bias by Regularizing Score Differences, Recsys 2022 の話をしようと思います。

発表資料 -> https://github.com/tmiya/tmiya.github.io/blob/main/paper_reading/20230719.pdf

fig_000

当日出た質問ですが、論文にちゃんと答えが書いてありました。評価値を +a, -a の2値化する働きがあると。