Few-Shot Continual Learning: Approaches and Future Directions

Few Shots Continual Learning (FSCL) is very challenging task, where the model has to continual learn with very few examples. This situation happens in many scientific and engineering domain. For example, an intelligent based system has to give diagnosis of disease and simultaneously learn from new unknown samples. Many researchers basically studied FSL and CL for two different objectives. FSL objective is to learn new task with very few examples, by learning transferable knowledge of some large base dataset and generalize for new task. CL objective is a sequential learning to mitigate catastrophic forgetting. Very less work has been contributed for FSCL as critical and challenging task. In this paper we have contributed intensive theoretical study on FSCL on different approaches to achieve FSL and CL together. Further we gave challenges and limitations associated with FSCL approaches with the potential applications of FSCL.