Towards Knowledgeable Supervised Lifelong Learning Systems

This paper co-authored by Diana Benavides-Prado offers a potential solution to challenges associated with machine learning systems that learn sequentially.  The authors propose a framework for long-term learning systems (Proficiente), which relies on transferring knowledge across tasks in two directions using Support Vector Machines (SVM).  Proficiente makes it possible to transfer knowledge acquired from a previous task forward to a new target task and to transfer knowledge acquired from recent tasks backward to refine knowledge acquired from previous tasks.  The authors demonstrate that transferring selected knowledge forward and backwards has the potential to encourage learning systems to become more knowledgeable while observing tasks sequentially.

Citation:
Benavides-Prado, D., Koh, Y. S., & Riddle, P. (2020). Towards Knowledgeable Supervised Lifelong Learning Systems. Journal of Artificial Intelligence Research, 68, 159-224.