Lifelong Event Detection with Embedding Space Separation and Compaction

arXiv:2404.02507v1 Announce Type: new
Abstract: To mitigate forgetting, existing lifelong event detection methods typically maintain a memory module and replay the stored memory data during the learning of a new task. However, the simple combination of memory data and new-task samples can still result in substantial forgetting of previously acquired knowledge, which may occur due to the potential overlap between the feature distribution of new data and the previously learned embedding space. Moreover, the model suffers from overfitting on the …

abstract acquired arxiv combination cs.cl data detection detection methods embedding event however knowledge memory samples simple space type