＜＜＜「@」を「__AT__」に置き換えています＞＞＞

From: Kenji Fukushima
Date: Tue, 29 Jan 2019 03:16:13 +0900
To: sg-l__AT__yukawa.kyoto-u.ac.jp
Subject: [Sg-l:4004] Fwd: [statphys:05540] 知の物理学研究センター セミナー ipi seminar [Mar. 7, Dr. Shotaro Shiba Funai]

素粒子論グループの皆さま：

東京大学「知の物理学研究センター」のセミナー案内を転送させて頂きます。

福嶋健二

---------- Forwarded message ---------
MLの皆様

東京大学の竹内一将と申します。
東京大学大学院理学系研究科で2018年12月に発足しました
知の物理学研究センター（iπ）主催のセミナー案内をさせて頂きます。
ご関心をお持ちの方々のご来聴を歓迎いたします。

Date:
Mar. 7 (Thu) 10:30-12:00

Place:
Room 913, Faculty of Science Bldg. 1,
Hongo Campus, The University of Tokyo
東京大学 本郷キャンパス 理学部1号館 913室

Speaker:
Dr. Shotaro Shiba Funai (OIST)

Title:
Renormalization, Thermodynamics, and Feature Extraction of Machine Learning

Abstract:
Recently the machine learning has been applied to data analysis in
various research fields. The methods of the machine learning are roughly
classified by supervised learning and unsupervised learning. In the
latter we train a machine so that it can reconstruct given dataset, then
the machine seems to extract features of the dataset. Since extraction
of feature resembles the coarse-graining, many researchers naively
consider it is closely related to the renormalization.

In this talk, however, I'd like to show that the feature extraction of
the machine learning has a clear difference from the renormalization.
For the training, we use the images of the spin configurations in Ising
model, since we know well about its renormalization group (RG) flow. We
generate the flow of images by iterative reconstructions of the machine,
and compare it with the RG flow. As a result, we find the reconstruction
flow shows some relations with renormalization and thermodynamic property.

http://www.phys.s.u-tokyo.ac.jp/wp-content/uploads/2019/01/ipi-seminar-190307.pdf

※英語によるセミナーとなります。
* Seminar will be given in English.

※会場へのアクセスはこちらをご覧ください。
https://www.s.u-tokyo.ac.jp/ja/map/map02.html
* Access
https://www.s.u-tokyo.ac.jp/en/map/map02.html

※知の物理学研究センターについて
http://www.phys.s.u-tokyo.ac.jp/lp/ipi/
* About Institute for Physics of Intelligence (iπ)
http://www.phys.s.u-tokyo.ac.jp/en/lp/ipi/