Long life learning system for document understanding

We present long life learning (LLL) system for understanding and processing administrative documents. Special attention was devoted to document classification and information extraction. These two modules represent the most significant part of LLL document understanding system. When changes occur in the document layout or when a novel class of documents appears, the system can adapt to these modifications by running auto-learning procedure. The system does not require a large training data set for creating the initial knowledge. Under specific conditions, it is possible to run the system without preliminary model training. The system will start without knowledge and continuously build and adapt necessary models with each document being processed from the input stream. Platform can process and effectively incorporate feedback from the user into already accumulated knowledge. The proposed solution is comparable to the concurrent systems known from the literature and in some respects even more innovative and appropriate to use in practice. Of course, to achieve accuracy close to a human user much more time, resources and common efforts of all dedicated research groups is needed.