A Survey: Navigating the Landscape of Incremental Learning Techniques and Trends

In traditional machine learning, model training usually involves training the model from scratch using the entire dataset. However, this approach often leads to catastrophic forgetting, i.e., the model rapidly and significantly forgets previously learned knowledge while training on samples from new tasks or new data distributions, when dealing with streaming data. Incremental learning, a particular machine learning paradigm, has drawn widespread attention as it allows the model to continuously learn from new data without forgetting the previously acquired knowledge. This paper comprehensively reviews existing incremental learning methods. The main contents of this survey include: the scenarios of incremental learning, a systematic review of incremental learning, a summary of performance evaluation metrics, and a discussion of open problems and future research directions in the area of incremental learning.