Title
Document Analysis and Machine Learning ... beyond simply fancy classification?
Name
Bart LamiroyAbstract
Advances in Document Analysis and Recognition over the last decade have been dominated by various Machine Learning breakthroughs and there is a current tendency towards YADLA (Yet Another Deep Learning Architecture/Approach) that seemingly solves a previously unachievable recognition or classification task, or improves upon the state of the art in a more or less impressive way. Looking beyond the strict DAR domain fancy commodity tools like mobile Google Translate that captures and translates text on the fly, or the same company's earbuds that allegedly provide a Babel fish-like service are proof that Machine Learning has significantly contributed to solving general recognition problems.
These achievements, however impressive they are, tend to be "just" fancy and well thought out implementations of fixed context classification tasks. Documents, however, tend to be vectors of human communication, and therefore intrinsically embed multiple levels of possible interpretation. Interacting with documents (from a human perspective) involves constant switching between various levels of interpretation and abstraction, often guided by the context of use. Does the current state-of-the-art in Document Analysis and Recognition address this topic? Can current Machine Learning classifiers be used to leverage progress in this domain? What are we lacking in our toolboxes to achieve automated interaction with documents?
Short Bio
Bart Lamiroy is a permanent faculty member at the Université de Lorraine, in Nancy, France, and member of the associated LORIA research lab. He was a visiting scientist at Lehigh University from January 2010 to July 2011. He has a broad experience in Machine Perception. Over the years, his research topics have ranged from Content Based Image Retrieval over Visual Servoing to Document Analysis. In 2007-2009 and 2016-2017, he was chair of the Computer Science and IT Department at the École des Mines de Nancy, France, of which he is now Deputy Dean of Studies. Before that he was a research contractor at INRIA, having obtained his Ph.D. in computer vision at the Institut National Polytechnique de Grenoble, France in 1998. He received his bachelor’s degree in applied mathematics in 1993. He currently serves as chair of the International Association for Pattern Recognition's TC-10 Committee on Graphics Recognition and is Associate Editor for the International Journal of Document Analysis and Recognition.