メニュー
| 日時 | 2020.11.5（10：00～12：00） |
|---|---|
| 会場 | |
| 講演タイトル | ※本セミナーはZoomにてご参加いただけます。 ライブのURLは当日朝に以下サイトでアナウンスされます。 https://sites.google.com/view/apsipa-jp 現在YouTubeで配信中です。 https://youtu.be/0cWYrD6vkJ4 ----------------------------------------------------------------------------------------------------------------------------- 10:00〜11:00 ◆Dr. Masahito Togami (LINE Corporation) "Deep speech source separation considering spatial model" 〈要旨〉 Deep neural network (DNN) is a powerful modeling tool for speech source spectrum. Although the DNN is an important technique, we believe that combination of the DNN with signal modeling based approaches, e.g., spatial model, is quite important. The signal models based on solid scientific knowledge can be expected to compensate for performance degradation when there is little data for learning or when there is mismatch between the learning environment and the evaluation environment. In this presentation, I will present several modern speech source separation techniques followed by our deep speech source separation which is combined with spatial modeling. 〈略歴〉 Masahito Togami received B.E., M.E., and Ph.D. degrees in aerospace engineering from the University of Tokyo in 2000, 2002, and 2011, respectively. He started artificial intelligence research of Hitachi's communication robot, EMIEW1-3. He received several awards such as the Awaya Award in 2009, the Itakura Award in 2010 from the Acoustical Society of Japan (ASJ), and the 29th TELECOM System Technology Award from the Telecommunications Advancement Foundation in 2014. He is a member of the ASJ and the JSAI and a senior member of the IEEE. He is a Board Member of the JSAI. He is also a sponsorship Co-Chairs of APSIPA2021. He joined LINE Corporation in 2018. He is now the manager and a principal researcher of speech team in LINE Corporation. He wrote ``Speech source separation with Python'' in japanese this year. 11:00〜12:00 ◆Dr. Antonio Ortega University of Southern California (U.S.A.), Department of Electrical Engineering "DeepNNK: A polytope interpolation framework for graph-based neural network analysis" 〈要旨〉 Modern machine learning systems based on neural networks have shown great success in learning complex data patterns while being able to make good predictions on unseen data points. However, these methods are hindered by time consuming model selection and lack of predictive explainability, especially in the presence of adversarial examples. In this talk, we take a step towards better understanding of neural networks by introducing a local polytope interpolation method. The proposed Deep Non Negative Kernel regression (NNK) interpolation framework is non parametric, theoretically simple and geometrically intuitive. We present the NNK graph construction, demonstrate instance based explainability and develop a method to identify models with good generalization properties using leave one out estimation. 〈略歴〉 Antonio Ortega is a professor of Electrical and Computer Engineering at the University of Southern California (USC). He is a Fellow of the IEEE and EURASIP, and a member of ACM and APSIPA. He is the Editor-in-Chief of the IEEE Transactions of Signal and Information Processing over Networks and recently served as a member of the Board of Governors of the IEEE Signal Processing Society. He has received several paper awards, including the 2016 Signal Processing Magazine award. His recent research work is focusing on graph signal processing, machine learning, multimedia compression and wireless sensor networks. |
| 言語 | 英語 |
| 対象 | どなたでも、ご参加いただけます。 |
| 共催 | Asia-Pacific Signal and Information Association Japan Chapter IEICE Signal Processing Symposium（電子情報通信学会信号処理シンポジウム） グローバルイノベーション研究院 ライフサイエンス分野 田中チーム 卓越大学院プログラム |
| お問い合わせ窓口 | グローバルイノベーション研究院 工学研究院 田中 聡久 e-mail: tanakat ( ここに@ を入れてください） cc.tuat.ac.jp |
このページの上部へ