Fig. 2From: JCBIE: a joint continual learning neural network for biomedical information extractionThe framework and component variations for jointly learning NER and RE. a The overall framework. b Single-head classifier. c Multi-head classifier. d Hard-parameter sharing encoder. e Soft-parameter sharing encoder. f no-parameter sharing encoder. g Continual learning multi-datasets. h Multi-corpora learning. i Continual multi-corpora learning. Figure indices with \(\bigstar\) (c, f, i) denote the proposed methods in JCBIE. The same components have the same color. SP denotes entity span; ET denotes entity type; CLS denotes classifier; ADE, DDI, and CPR denote different datasets, containing different entity types and relation typesBack to article page