The scope of this paper is to give an outline of future Postal Address Readers whose prominent characteristic is their learning capability, which can be expressed as learning from letters. In contrast to current reader systems with theirpredefined and frozen parameters and knowledge bases, they will be able to adapt themselves automatically to the changing characteristics of mail mix and writing conventions. The structure of such a system and the process of online learning is explained.Three cases of adaptiveness are treated: instantaneous, short and long term adaptiveness, i.e. learning from the same mail piece, from the immediately previous one and from all previous mail pieces. Such an adaptive system is beneficial for both thecustomer and the supplier, since it offers for the first time an optimized solution for each site at any time and it helps save a lot of engineering effort for tedious local system improvements.
展开▼
