Abstract
The talk introduces some of our work on cross-modal understanding of language and images, starting from exploring how humans turn language into meaning. We start from the relationship between vision and language, and use 30 million pairs of images and text generated by the Internet to complete a self-supervised task for the largest Chinese general-purpose graphical pre-training model, Wudao ∙ Wenlan, and thus go on to initially explore the possibility of AI learning language in a multimodal environment.
Prof.Ruihua Song
She joined Microsoft Research Asia in 2003 and has been working on information retrieval and mining, recommendation and text generation. Her proposed web page chunking and importance prediction was nominated for the best paper at the international World Wide Web conference WWW 2004. Her proposed generic method for extracting subject content from different websites is at the core of the reading function of Microsoft browser after IE 11. In May 2017, Zhanlu Culture published the first AI poetry book in human history, Sunlight Lost the Glass Window, in which all the poems were generated by the very algorithm she led to develop. She then joined Microsoft Ice as Chief Scientist until July 2020, and in September 2020, Dr. Song joined Renmin University of China's High Tide Institute of Artificial Intelligence as a long-tenured Associate Professor. Her recent research interests include text creation for artificial intelligence, multimodal understanding of natural language, and multimodal dialogue systems. Dr. Ruihua Song is an internationally influential scientist who has published more than 80 academic papers in international conferences and journals and holds more than 25 international patents. She serves as a domain chair, senior reviewer and reviewer for several leading international conferences (SIGIR, CIKM, EMNLP, etc.), editorial board member of international journals (Information Retrieval Journal and Foundations and Trends in Information Retrieval). 2021 She will be the chair of SIGIR Short Papers.
