List of Keynote Talks

✏️ Machine Learning for Creative Workflow

Time: December 14, 13:00 - 14:00

Keynote Speaker: Kota Yamaguchi

Kota Yamaguchi is a research manager at CyberAgent AI Lab. He currently works on research and development of computer vision and machine learning techniques for creative workflow automation. He was previously an assistant professor at Tohoku University from 2014 to 2017. He received a Ph.D. degree in Computer Science from Stony Brook University in 2014. He received an MS in 2008 and a BE in 2006, both from the University of Tokyo.


Recent advancement of machine learning techniques transforms how creators design and finish up their work. In this talk, I present ongoing research efforts at CyberAgent in bringing the machine learning techniques to intelligently assist graphic designers in the digital advertising industry. The data structure around creative workflow is characterized by a vector graphic format that precisely describes the high-level multi-modal structure of the final visual presentation, which poses several challenges to machine learning models from data representation to performance evaluation. We will discuss how we approach design tasks such as text de-rendering from raster images and unsupervised document generation.

✏️ Connecting the Dots: Digital Humanities and Historical Big Data Research for Japanese Culture

Time: December 15, 13:00 - 14:00

Keynote Speaker: Asanobu Kitamoto

Asanobu Kitamoto earned his Ph.D. in electronic engineering from the University of Tokyo in 1997. He is now the Director of the Center for Open Data in the Humanities (CODH), Joint Support-Center for Data Science Research (DS), Research Organization of Information and Systems (ROIS), Professor of the National Institute of Informatics, and SOKENDAI (The Graduate University for Advanced Studies). He has developed various data-driven science approaches in fields such as the humanities, earth sciences, and disaster management. He has released databases and software as academic research platforms with a few million users from academia and society. He is also working on a trans-disciplinary collaboration to promote open science. He has received awards such as Jury Recommended Works (Art Division) from Japan Media Arts Festival, Yamashita SIG Research Award from the Information Processing Society of Japan (IPSJ), Best Paper Award from IPSJ SIG Computers and the Humanities Symposium, Academic Award (Research Paper) from Japan Society for Digital Archive, and Good Design Award.


Data-driven approaches such as machine learning and multimedia technology can accelerate research on Japanese culture from a historical perspective. Furthermore, thanks to a widespread movement toward open science, such as open data and open source, research on Japanese culture has finally entered into a big data era. However, humanities data is full of interpretation and meaning, has a complex structure with diversity, and requires implicit knowledge not explicitly described inside the humanities data. Hence we do not expect that a technological silver bullet, such as AI, can solve all the problems at once. Instead, we employ a connect-the-dot approach by developing and connecting tools and datasets to answer humanities research questions for understanding Japanese culture. For that purpose, we carefully gather evidence from many sources in the past, create structured data with human-machine collaborations, and integrate them as linked data to draw a bigger picture of Japanese culture looking backward from the present. The talk will introduce our activities in the ROIS-DS Center for Open Data (CODH) in the Humanities to tackle these challenges as digital humanities and historical big data research. The talk will include machine learning for the recognition of Kuzushiji (Japanese historical cursive scripts), art history research using IIIF (International Image Interoperability Framework), and computer vision-based differential reading for diachronic transcription. We also discuss the importance of domain knowledge and collaboration with domain experts to ask meaningful research questions beyond simplistic metric-based evaluations.