Cross-Lingual Image Caption Generation
Automatically generating a natural language description of an image is a fundamental problem in artificial intelligence. Recently, image caption generation made significant progress by deep learning based methods and large image caption datasets. However, there are only a few multi-lingual datasets, the progress was restricted in English caption generation. In this talk, I will describe a simple method to improve the performance of image caption generation in non-English language (Japanese, in this case) by taking advantage of relatively large English image caption dataset.
Takashi Miyazaki is a research engineer in Yahoo Japan. He received the PhD in Computer Science from Keio University. He then studied neuroscience as a post-doctoral researcher at National Institute for Physiological Sciences and at the University of Tokyo. He is currently working on vision and language problems using deep learning methods.