6/21に第5回学生ゼミが開催されました．
今回はB4の嶋野くんから，5章の畳み込みニューラルネットワーク（CNN）について発表しました．
基本的な畳み込みの計算から多様な形式のCNNを紹介していただきました．分厚い章で，お疲れ様でした．
同変性と不変性のところ，Group Convolutional Neural Networksという，回転や反転まで同変性を持たせるための研究について一緒に読んでみました．私は群論に関する知識が足りなくて，苦戦しました．
後半は，OBの稲福くんにオンラインで参加してもらい，彼がB4時にCNNを勉強して研究で用いた話しを伺いました．物体検出やセグメンテーションなどの手法はやはり面白いし，幅広く使えますね．
次回は第12章の生成モデル，自己符号化器や変分自己符号化器について扱います．
最近は梅雨の季節で，雨が降ったりやんだり，暑かったりで，皆様くれぐれもご自愛くださいね．
The 5th student seminar was held on 21st June.
Mr Shimano (B4) gave a presentation on Chapter 5, convolutional neural networks (CNN).
It was an abundant chapter, introducing everything from basic convolutional computation to various forms of CNNs.
In the section on equivariance and invariance, we read together about Group Convolutional Neural Networks, a study to keep the equivariance about rotations and flips. I struggled with it because I lacked knowledge about group theory.
Besides, Mr. Inafuku, an alumni of our lab, joined us online. He shared about how he studied CNNs during his B4 and used them in his research. Methods such as object detection and segmentation are interesting and widely used.
Next seminar, we will deal with generative models, such as self-encoders and variational self-encoders in Chapter 12.
It’s the rainy season these days. Sometimes it rains and sometimes it’s hot. Please take care.
投稿者｜邱