Tutorials - チュートリアル
MIRU2014チュートリアルセッションを7月28日（月）に開催いたします．
MIRU2014のチュートリアルセッションは２部構成
今回のチュートリアルセッションは，画像の認識・理解に関する研究に従事する研究者がいま学びたい技術トピックを取り扱う「技術チュートリアル」４セッションと，「画像の次」に我々が向かうべき方向を議論する「特別チュートリアル」の２部構成です．
日時
7月28日（月）11:00〜18:15
場所
MIRU会場（岡山コンベンションセンター）
フロア図（ http://www.mamakari.net/facility/ )
参加費
MIRU本会議参加登録料に含まれます
MIRU2014 Tutorials will be held at July 28 (Mon) in the same venue as the main conference. The tutorials mainly consists of two parts, i.e. four Technical tutorials providing knowledge from introduction to application on hot research topics for MIRU community and one Organized (special) tutorial discussing about frontier lying beyond image-based researches. The registration for all tutorials is included MIRU2014 registration.
タイムテーブル - Time Table
＊チュートリアルは日本語で行われます．
＊All tutorials are held in Japanese.
＊発表資料事前公開について：資料へアクセスするためのアドレス及びパスワードは，MIRU2014に事前参加登録された方々にメールでお知らせいたします．
技術チュートリアル1 - Technical Tutorial 1
「Deep Learning ～使いこなすために知っておきたいこと～」
山下隆義（中部大学）
本チュートリアルでは画像認識や音声認識などにおいて注目されているディープラーニングについて，基本的な手法から最新の手法まで紹介する．また，ディープラーニングはパラメータ数が膨大であり，使いこなすまでに大変労力がかかる．どのようなパラメータがあり，どう作用するか等の使いこなすためのノウハウや，世の中にあるツール等の紹介も行う
How to master Deep Learning
Takayoshi Yamashita (Chubu University)
In this tutorial, I will introduce the method from basic techniques to latest ones about deep learning that has archived state of the art performance in speech recognition and image recognition. Deep learning has enormous the number of parameters and it takes long to time to master and obtain the know-how. I will also introduce the behavior about the parameters and the efficient tools to implement or evaluate the deep learning.
2002年奈良先端大科学技術大学院大学 博士前期課程修了．
同年オムロン株式会社入社（～2014年）．
2009年中部大学大学院 博士後期課程修了．
2014年4月より中部大学工学部 講師．
顔および人検出などの人を理解する画像認識，物体追跡などの動画像解析の研究に従事．
Takayoshi Yamashita received his MS degree in Engineering in 2002 at NAIST, and PhD degree in Engineering in 2009 at Chubu University.
He worked OMRON from 2002 to 2014.
He is a lecturer at Chubu University since 2014.
His research interests include face and human detection, object tracking and machine learning.
技術チュートリアル2 - Technical Tutorial 2
「映像解析ベンチマークTRECVID: 「ワイルド」な映像を使った意味解析，物体検索，イベント検出と性能評価」
佐藤真一（国立情報学研究所）
TRECVIDは，情報検索の分野から生まれてきた映像解析・検索のための競争型ワークショップであり，提供されるデータセット，正解データ，タスクは標準 的なベンチマークとして広く利用されている．しかし，情報検索から生まれてきたこともあり，PASCAL VOCやImageNetなどコンピュータビジョン研究者らが作り出してきたベンチマークデータとは異なる面も多い．にもかかわらず，昨今，コンピュータビジョン研究者らの間でもTRECVIDを使った研究が盛んになってきている．本チュートリアルでは，TRECVIDの概要に触れた上で，他のベンチマークとの違いは何か，TRECVIDによるベンチマークの利点は何か，等の点について迫りたい．
TRECVID the benchmark for video analysis
-- semantic analysis, object retrieval, event detection, and performance evaluation using "wild" video dataset
Shin'ichi Satoh (National Institute of Informatics)
TRECVID was initiated by information retrieval community and has been world wide competitive benchmark campaign for video analysis and retrieval. The video dataset, ground truth, and tasks provided by TRECVID are now recognized as de facto benchmark and are widely used by researchers. However, partly because of its origin in information retrieval community, there are different aspects between TRECVID and other benchmarks created by computer vision community such as PASCAL VOC and ImageNet. Despite this fact, TRECVID is now becoming popular even in computer vision community. This tutorial will first brief TRECVID, followed by the discussion on the difference between TRECVID and other benchmarks, pros and cons of TRECVID, and so on.
1987東京大学工学部電子工学科卒．1992同大大学院工学系研究科情報工学専攻 博士課程了．同年学術情報センター助手．1998同助教授．2000国立情報学研究所助教授．2004同教授．現在に至る．1995から1997まで，米国カーネギーメロ ン大客員研究員としてInformedia映像ディジタルライブラリの研究に従事．工博．
Shin'ichi Satoh received his BE degree in Electronics Engineering in 1987, his ME and PhD degrees in Information Engineering in 1989 and 1992 at the University of Tokyo. He joined National Center for Science Information Systems (NACSIS), Tokyo, in 1992. He is a full professor at National Institute of Informatics (NII), Tokyo, since 2004. He was a visiting scientist at the Robotics Institute, Carnegie Mellon University, from 1995 to 1997. His research interests include image processing, video content analysis and multimedia database. Currently he is leading the video processing project at NII.
技術チュートリアル３ - Technical Tutorial 3
スパースモデリングの基礎
日野 英逸（筑波大学）
スパース性が種々の現象のモデリング原理として有用であることが知られるようになって久しい．統計学における急速な理論的発展と，信号処理を始めとする工学分野での爆発的な利用，そして脳神経科学やバイオインフォマティクス，天文科学にまで及ぶ理学的原理としての展開はとどまるところを知らない．画像処理分野に限っても，画像の分離，修復，ノイズ除去，顔画像による認証，超解像等，その応用範囲は多岐にわたる．
本チュートリアル講演では，スパースモデリングの工学的応用のための基礎知識として，正則化回帰を中心とするスパースモデリングの基礎と，具体的にスパースモデリングで多用される推定アルゴリズムを幾つか紹介する．特に機械学習やパターン認識分野における応用例を紹介し，スパースモデリングの今後の展望について，講演者の最近の研究と，関連した話題を紹介する．
Introduction to Sparse Modeling
Hideitsu Hino (University of Tsukuba)
The notion of sparseness has been known to be a useful modeling principle of various phenomena for a long time. Principle of sparse modeling is supported by rapid theoretical development in statistics, widespread use in engineering and science. The application of sparse modeling covers a broad range of scientific fields from neuroscience to astronomical science.
In this tutorial talk, as basic knowledge for the engineering applications of sparse modeling, we first introduce basic mathematics for sparse modeling, particularly focusing on regularized regression modeling and some useful algorithms. Then, we introduce application examples in the field of machine learning and pattern recognition. Finally, we introduce our recent works and make mention of the future prospects of the sparse modeling.
平15 京大工学部卒，平17 同大院情報学研究科 修士課程了． 同年より日立製作所勤務．平22 早大院博士課程了． 平25 筑波大学助教，現在に至る．博士(工学). 機械学習，データ解析の研究に従事．
Hideitsu Hino received his Bachelor's degree in engineering in 2003, and Master's degree in Applied Mathematics and Physics in 2005 from Kyoto University, Japan. He joined Hitachi's Systems Development Laboratory and worked as a research staff from April 2005. He earned Doctor's degree in engineering in 2010 from Waseda University. From April 2013, he is working as an Assistant Professor in University of Tsukuba. His research interest include machine learning and data analysis.
技術チュートリアル４ - Technical Tutorial 4
Transfer Learning: Theory and Applications
Osamu Hasegawa (Tokyo Institute of Technology)
In this talk, we would like to introduce theories and applications of “Transfer Learning”. There are enormous numbers of objects and tasks in the real-world, and observed data from them contain much unpredictable noise. For example, if we want a computer to recognize a microwave oven, by the conventional approach, we would need to collect large numbers of various microwave oven images for learning and let the computer learn for long time.
On the other hand, in the Transfer Learning approach, we let the computer recognize the microwave oven as a square white object that has a dial and a buzzer that sounds.” In other words, we let the computer learn basic knowledge about the real-world such as shapes and colors, sounds, hardness, and weight beforehand, and then, let the computer recognize objects by a combination of such basic knowledge. Theoretically, by this approach, the computer can recognize large numbers of objects in the real world by learning fewer basic and fundamental facts.
The other important advantage of the Transfer Learning approach is that the computer can guess unknown objects. Please imagine that you see a product with a description written in unknown language, you would not understand exactly what it is. However, I think you could categorize what it is, for example,“maybe, a kind of orange juice”. This is because all of you have learnt basic and fundamental knowledge from your daily life and can transfer such knowledge to the product. In other words, everyone acquires basic and fundamental knowledge from birth, such as weight of various things, colors and textures, sounds, smells, tastes, through everyday life. Thanks to this knowledge, by the time we are adults, we can robustly recognize the real world.
Lastly, in our talk, we plan to show a demonstration of Transfer Learning controlled by SOINN as one of the applications of it.
Dr. Osamu Hasegawa is a Japanese researcher in the fields of computer science and robotics at Tokyo Institute of Technology. As the Principle Investigator (PI) of the Imaging Science and Engineering Laboratory, he is leading a research group to further develop an innovative learning mechanism, "SOINN". In addition to being a PI, he is an appointed Associate Professor in the Department of Computational Intelligence and Systems Science, a position that he has held since 2002.
MIRU2014 チュートリアル担当：大山 航 （三重大学）