Evolving Connectionist Systems For On-Line Pattern Classification Of Multimedia Data

The paper develops further the concept of evolving connectionist systems, and one particular model - evolving fuzzy neural networks, that are applied on pattern classification tasks of multimedia data. The evolving systems learn in an on-line, life-long learning mode and adapt to the new data. This mode is crucial when the system is required to adapt quickly to new data and be able to generalize immediately afterwards. These tasks are typical for processing of multimedia data, such as adaptive speech recognition and adaptive video data processing that are presented in the paper.