I am Satoshi Tsutsui, a Ph.D. student at School of Informatics and Computing, Indiana University. I started my Ph.D. life at 2015 Fall. I am advised by Prof. David Crandall and Prof. Ying Ding.
While it is somewhat over-advertised these days, deep learning accelerates the progress of AI. Hence I follow the latest progress, and sometimes play with the cutting edge papers by reproducing them. I use chainer and tensorflow. You can find implementations in my github. I also have several posts on my blog.
Can we teach computers how to see as we do? This is very challenging research area but also very fun domain to work on. Recently computer vision suddenly started to work due to deep learning. I like image understanding problems such as object recognition, object detection, and caption generation. In addition, I am recently intreted in generative adversarial networks.
Mining figures in scientific papers has recently become an interesting area of research. In computer vision, some people started to work on parsing figures appearing in research papers or textbook. In information science, a figure oriented literature mining is recently proposed. This is called Viziometrics in contract to bibliometrics, which is mainly focused on text.
What I want to do can be stated in two phases. First is to teach computers understand figures in human level. Can we understand a figure and perform reasoning based on the understanding? This is computer vision work. After I develop computer vision techniques, I want to apply the techniques to large collections of papers, then want to find interesting knowledge. I believe that figures can tell interesting stories of academic fields.
The number of medical papers is increasing in exponential order. This is a real issue for people working on medical domain but also a chance to to build AI that can read papers, and ultimately discover new knowledge in a data driven way. I collaborate with researchers on Alzheimer’s Disease to try this idea.
Techniques to be used will be information extraction, automatic knowledge base construction, and graph/network mining.
The state of the art artificial intelligence is dominated by data driven approach. However, I still believe knowledge is also important to build intelligent computer. Data-driven driven approach cannot let computers understand the world in semantic level. Knowledge based approach need to be integrated with data-driven driven approach. In that sense, Semantic Web technologies can provide a way to equip computers knowledge with semantic understanding.
I learned semantic web technologies in my undergraduate research program. I extended DBpedia using list structure in Wikipedia pages. If you are interested, please refer to this unpublished paper.