How much Position Information Do Convolutional Neural Networks Encode? | OpenReview

How much Position Information Do Convolutional Neural Networks Encode? Sep 25, 2019 Blind Submission readers: everyone Show Bibtex Keywords: network understanding, absolute position information TL;DR: Our work shows positional information has been implicitly encoded in a network. This information is important for detecting position-dependent features, e.g. semantic and saliency. Abstract: In contrast to fully connected networks, Convolutional Neural Networks (CNNs) achieve efficiency b

2 mentions: @icoxfog417
Date: 2020/01/15 00:51

Referring Tweets

@icoxfog417 CNNは位置情報を学習しているか、学習しているとしたらどこ情報があるか検証した研究。VGG/ResNetの中間特徴から、X/Y方向等のピクセル強度勾配を予測できるか検証している(入力画像(犬or猫等)とは独立の出力を設定)。位置情報は深い層にあり、paddingが学習の肝との結果。 t.co/IhesUHkrat

Related Entries

Read more GitHub - rlworkgroup/metaworld: An open source robotics benchmark for meta- and multi-task reinforce...
0 users, 1 mentions 2019/09/14 03:48
Read more Weekly Machine Learning #149 | RevueRevue
1 users, 1 mentions 2019/11/09 06:51
Read more SampleRNN: An Unconditional End-to-End Neural Audio Generation Model | OpenReview
0 users, 1 mentions 2020/03/06 00:52
Read more Connected Papers | Find and explore academic papers
0 users, 1 mentions 2020/06/11 14:21
Read more ACL2020Program - Google Sheets
0 users, 2 mentions 2020/07/05 12:58