Lifelong machine learning: a paradigm for continuous learning

摘要

Although LML has been around for more than 20 years, not a great deal of research has been done so far. One reason could be that the ML research in the past 20 years focused on statistical and algorithmic approaches. LML typically needs systems approaches. However, as statistical machine learn-ing becomes mature and researchers realize its limitations, LML will become more and more important. It is probably safe to say that without the LML capability to accumulate learned knowledge and learn new tasks with the help of the past knowledge in a self-motivated manner, we will not be able to build a truly intelligent system. We can only solve problems in very narrow domains.