講演名 2021-09-17
Identifying Design Problems of Presentation Slides using a Bimodal Neural Network
易 聖舟(東大), ?上 純一郎(ルバート), 山崎 俊彦(東大),
PDFダウンロードページ PDFダウンロードページへ
抄録(和) Although millions of presentation slides are created every day in business and academia, there are only a limited number of support systems that are helpful to assess the slides. In this study, a bimodal neural network, using visual and structural features, is proposed to identify the design problem of presentation slides. For such a purpose, over two thousand slides were collected for training the model. We summarized ten checkpoints, which are common problems in slide design. The dataset faces an imbalanced distribution, because only a small part of the samples had corresponding design problems. To address the class imbalance issue, several sampling methods are applied to improve the prediction performance. Furthermore, we also use transfer and multi-task learning to enhance the bimodal neural network. The optimal combination of these machine-learning methods helps the proposed network achieve an average accuracy of 81.79%.
抄録(英) Although millions of presentation slides are created every day in business and academia, there are only a limited number of support systems that are helpful to assess the slides. In this study, a bimodal neural network, using visual and structural features, is proposed to identify the design problem of presentation slides. For such a purpose, over two thousand slides were collected for training the model. We summarized ten checkpoints, which are common problems in slide design. The dataset faces an imbalanced distribution, because only a small part of the samples had corresponding design problems. To address the class imbalance issue, several sampling methods are applied to improve the prediction performance. Furthermore, we also use transfer and multi-task learning to enhance the bimodal neural network. The optimal combination of these machine-learning methods helps the proposed network achieve an average accuracy of 81.79%.
キーワード(和) Presentation Slide / Class Imbalance / Multi-Task Learning / Transfer Learning
キーワード(英) Presentation Slide / Class Imbalance / Multi-Task Learning / Transfer Learning
資料番号 MVE2021-12
発行日 2021-09-10 (MVE)

研究会情報
研究会 MVE
開催期間 2021/9/17(から2日開催)
開催地(和) オンライン開催(国士舘大学でのハイブリッド開催から変更)
開催地(英) Online
テーマ(和) リモートエクスペリエンスの魅力(「リモートワーク」「リモートコラボレーション」「ワーケーション」)、メディアエクスペリエンスおよび一般(魅力工学(AC)研究会協賛)
テーマ(英)
委員長氏名(和) 井原 雅行(理研)
委員長氏名(英) Masayuki Ihara(RIKEN)
副委員長氏名(和) 清川 清(奈良先端大)
副委員長氏名(英) Kiyoshi Kiyokawa(NAIST)
幹事氏名(和) 西口 敏司(阪工大) / 横山 正典(NTT) / 福嶋 政期(東大)
幹事氏名(英) Satoshi Nishiguchi(Oosaka Inst. of Tech.) / Masanori Yokoyama(NTT) / Shogo Fukushima(Univ. of ToKyo)
幹事補佐氏名(和) 磯山 直也(奈良先端大) / 原 豪紀(大日本印刷) / 後藤 充裕(NTT)
幹事補佐氏名(英) Naoya Isoyama(NAIST) / Takenori Hara(DNP) / Mitsuhiro Goto(NTT)

講演論文情報詳細
申込み研究会 Technical Committee on Media Experience and Virtual Environment
本文の言語 ENG
タイトル(和)
サブタイトル(和)
タイトル(英) Identifying Design Problems of Presentation Slides using a Bimodal Neural Network
サブタイトル(和)
キーワード(1)(和/英) Presentation Slide / Presentation Slide
キーワード(2)(和/英) Class Imbalance / Class Imbalance
キーワード(3)(和/英) Multi-Task Learning / Multi-Task Learning
キーワード(4)(和/英) Transfer Learning / Transfer Learning
第 1 著者 氏名(和/英) 易 聖舟 / Shengzhou Yi
第 1 著者 所属(和/英) 東京大学(略称:東大)
The University of Tokyo(略称:UTokyo)
第 2 著者 氏名(和/英) ?上 純一郎 / Junichiro Matsugami
第 2 著者 所属(和/英) 株式会社ルバート(略称:ルバート)
Rubato Co., Ltd.(略称:Rubato)
第 3 著者 氏名(和/英) 山崎 俊彦 / Toshihiko Yamasaki
第 3 著者 所属(和/英) 東京大学(略称:東大)
The University of Tokyo(略称:UTokyo)
発表年月日 2021-09-17
資料番号 MVE2021-12
巻番号(vol) vol.121
号番号(no) MVE-179
ページ範囲 pp.21-26(MVE),
ページ数 6
発行日 2021-09-10 (MVE)