Updated on 2022/05/25

写真a

 
Yasuda, Goki
 
Affiliation
Affiliated organization, Center for Data Science
Job title
Associate Professor(tenure-track)

Concurrent Post

  • Affiliated organization   Global Education Center

Research Institute

  • 2020
    -
    2022

    データ科学センター   兼任センター員

 

Research Areas

  • Theory of informatics

  • Intelligent informatics

  • Statistical science

Research Interests

  • Machine Learning

  • Statistical Learning Theory

  • Statistical Science

  • Information Theory

Papers

  • Asymptotoic Analysis of Classification in the Presence of Generalized Label Noise

    Proceedings of International Symposium on Information Theory and Its Applications    2018.10  [Refereed]

  • A Study of Asymptotic Evaluation of Prediction for Semi-supervised Learning

    G. Yasuda

    Waseda University    2017.02

  • Asymptotics of Prediction Based on a Consistent Solution of the Likelihood Equation for Semi-supervised Learning

    G. Yasuda, N. Miya, T. Suko, T. Matsushima

    IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences (Japanese Edition)   vol. J100-A ( 1 ) pp. 102 - 113  2017.01  [Refereed]

  • Analysis of Performance Gain from Unlabeled Data in Semi-supervised Learning

    Proceedings of International Symposium on Information Theory and Its Applications    2016.10  [Refereed]

  • Asymptotics of Bayesian Inference for a Class of Probabilistic Models under Misspecification

    Nozomi Miya, Tota Suko, Goki Yasuda, Toshiyasu Matsushima

    IEICE TRANSACTIONS ON FUNDAMENTALS OF ELECTRONICS COMMUNICATIONS AND COMPUTER SCIENCES   E97A ( 12 ) 2352 - 2360  2014.12  [Refereed]

     View Summary

    In this paper, sequential prediction is studied. The typical assumptions about the probabilistic model in sequential prediction are following two cases. One is the case that a certain probabilistic model is given and the parameters are unknown. The other is the case that not a certain probabilistic model but a class of probabilistic models is given and the parameters are unknown. If there exist some parameters and some models such that the distributions that are identified by them equal the source distribution, an assumed model or a class of models can represent the source distribution. This case is called that specifiable condition is satisfied. In this study, the decision based on the Bayesian principle is made for a class of probabilistic models (not for a certain probabilistic model). The case that specifiable condition is not satisfied is studied. Then, the asymptotic behaviors of the cumulative logarithmic loss for individual sequence in the sense of almost sure convergence and the expected loss, i.e. redundancy are analyzed and the constant terms of the asymptotic equations are identified.

    DOI

  • Asymptotics of MLE-based Prediction for Semi-supervised Learning

    Proceedings of International Symposium on Information Theory and Its Applications    2014.10  [Refereed]

  • Implementation and optimization of software 2Dto3D conversion for ARM

    Tse Kai Heng, Yuji Kawashima, Tatsuro Fujisawa, Goki Yasuda, Makoto Oshikiri, Takahiro Tanaka, Kaoru Matsuoka, Yoshihiro Kikuchi

    1st IEEE Global Conference on Consumer Electronics 2012, GCCE 2012     491 - 492  2012  [Refereed]

     View Summary

    3D imaging capable devices have been available on the market for a few years now but the number of 3D contents available for viewing is not growing fast enough to meet the demand. To compensate for such a problem, most 3D imaging devices are equipped with a 2Dto3D conversion function. In 2Dto3D conversion, depth estimation is essential in order to create pseudo 3D images. Often, such conversion involves complex computations and is difficult to process in a real-time manner. In this paper, we present the implementation of Toshiba's software 2Dto3D conversion solution, in particular, the depth estimation process on ARM processor. This could potentially allow real-time processing on embedded devices such as TVs, mobile phones and tablet computers. © 2012 IEEE.

    DOI

  • Asymptotics of Bayesian estimation for nested models under misspecification

    Nozomi Miya, Tota Suko, Goki Yasuda, Toshiyasu Matsushima

    2012 INTERNATIONAL SYMPOSIUM ON INFORMATION THEORY AND ITS APPLICATIONS (ISITA 2012)     86 - 90  2012  [Refereed]

     View Summary

    We analyze the asymptotic properties of the cumulative logarithmic loss in the decision problem based on the Bayesian principle and explicitly identify the constant terms of the asymptotic equations as in the case of previous studies by Clarke and Barron and Gotoh et al. We assume that the set of models is given that identify a class of parameterized distributions, it has a nested structure and the source distribution is not contained in all the families of parameterized distributions that are identified by each model. The cumulative logarithmic loss is the sum of the logarithmic loss functions for each time decision-, e. g., the redundancy in the universal noiseless source coding.

  • IN-LOOP FILTER USING BLOCK-BASED FILTER CONTROL FOR VIDEO CODING

    Takashi Watanabe, Naofumi Wada, Goki Yasuda, Akiyuki Tanizawa, Takeshi Chujoh, Tomoo Yamakage

    2009 16TH IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING, VOLS 1-6     1013 - 1016  2009  [Refereed]

     View Summary

    In this paper, a method of improving coding efficiency is proposed by using the Wiener filter as an in-loop filter. The Wiener filter can minimize the mean square error between the input image and the decoded image. However, errors of some pixels increase by filtering process. Since the filtered pixels are used for motion-compensated prediction, these errors are propagated to the subsequent images. The proposed method divides the decoded image into some fixed blocks, and decides whether to apply the filter for each block adaptively. As a result, by preventing the increase in errors after the filtering process, the coding efficiency can be improved. Experimental results show that the proposed method achieves bitrate reduction of up to 33.9% in Baseline Profile and up to 33.0 % in High Profile at the same PSNR compared to H.264.

    DOI

  • Extended Adaptive Filtering for Motion Compensated Prediction

    Proceedings of International Workshop on Advanced Image Technology    2006.01  [Refereed]

  • 拡張適応補間フィルタを用いた動き補償予測

    G. Yasuda, T. Chujoh

    Forum on Information Technology   FIT 2004 ( 3 ) pp. 223 - 224  2004.09  [Refereed]

  • 補助情報付情報源のベイズ決定理論に基づくユニバーサル符号化

    G. Yasuda

    Waseda University    2003

▼display all

Misc

  • 半教師付き学習における予測誤差に対するラベルなしデータの有効性に関する一考察

    大和田歩, 安田豪毅, 松嶋敏泰

    情報理論とその応用シンポジウム予稿集   37   pp. 288 - 293  2014.12

    Research paper, summary (national, other academic conference)  

  • 真の分布を含むとは限らない階層モデル族に対するベイズ推定の漸近評価

    宮希望, 須子統太, 安田豪毅, 松嶋敏泰

    情報理論とその応用シンポジウム予稿集   36   pp. 665 - 670  2013.11

    Research paper, summary (national, other academic conference)  

  • 半教師付き学習における一致推定量に基づく予測の漸近評価

    安田豪毅, 宮希望, 須子統太, 松嶋敏泰

    情報理論とその応用シンポジウム予稿集   36   pp. 659 - 664  2013.11

    Research paper, summary (national, other academic conference)  

  • Asymptotics of Bayesian prediction for misspecified models

    N. Miya, T. Suko, G. Yasuda, T. Matsushima

    IEICE Technical Report (IT)   111 ( 142 ) pp. 71 - 76  2011.07

    Research paper, summary (national, other academic conference)  

  • A study of coding for sources with nonstationary parameter

    G. Yasuda, R. Nomura, T. Matsushima

    IEICE Technical Report (IT)   101 ( 177 ) pp. 25 - 30  2001.07

    Research paper, summary (national, other academic conference)  

Specific Research

  • 様々な品質のデータからの機械学習に関する研究

    2019   須子統太, 堀井俊佑, 小林学, 松嶋敏泰

     View Summary

    本研究では,ラベルノイズを含むデータからの学習について,理論的な性能と実際に分類アルゴリズムによって得られる性能の比較,および潜在クラスによってラベルノイズが異なる場合の分類アルゴリズムの開発を行った.性能の比較では,理論的な性能の限界式の値を数値計算によって算出し,EMアルゴリズムによる分類アルゴリズムの性能と比較した.性能の比較から,その分類アルゴリズムによって理論的な性能の限界に十分近い性能が得られることが確認できた.また,分類アルゴリズムの開発については,人工データによる実験により,その有効性が確認された.これらの研究成果について,国内学会にて2件の発表を行った.

  • 半数付き学習における予測の性能評価

    2018   須子 統太, 小林 学

     View Summary

    本研究では,一般化ラベルノイズを含むデータからの学習について,理論と実験の両面から性能評価を行なった.一般化ラベルノイズを含むデータからの学習は,半教付き学習や,外れ値を含むデータからの学習等の様々な学習を包含しており,研究計画時よりも広い学習を対象とすることができた.理論的な性能評価では,学習データについての仮定を明示した上で,理想的なラベルの予測方法について性能限界を求めた.実験による性能評価では,ラベルノイズに関する確率分布が未知の場合について,分類アルゴリズムを提案し,その性能を確認した.これらの研究成果について,国際学会にて1件,国内学会にて2件の発表を行なった.

 

Syllabus

▼display all