Skip to content

Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases #35

@nocotan

Description

@nocotan

一言でいうと

自己教師あり表現学習によるパフォーマンス向上の謎を解くための数値実験を実行.

論文リンク

https://papers.nips.cc/paper/2020/file/22f791da07b0d8a2504c2537c560001c-Paper.pdf

著者/所属機関

Senthil Purushwalkam and Abhinav Gupta
(Carnegie Mellon University)

投稿日付(yyyy/MM/dd)

2020/12

概要

自己教師あり表現学習は多くのダウンストリームのタスクで純粋な教師あり学習を超えている.
近年のパフォーマンスの向上は,インスタンス分類モデルを学習し,各画像とそのデータ拡張版を単一のクラスのサンプルとして扱うことによっている.
論文では,そうしたパフォーマンス向上の謎を解くための数値実験を行う.

Screen Shot 2021-02-26 at 2 23 27

新規性・差分

  • 自己教師あり表現学習によるパフォーマンス向上の謎を解くための数値実験を実行.

手法

Screen Shot 2021-02-26 at 2 23 53

結果

Screen Shot 2021-02-26 at 2 24 02

Screen Shot 2021-02-26 at 2 24 13

Screen Shot 2021-02-26 at 2 24 21

Screen Shot 2021-02-26 at 2 24 30

コメント

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions