Biomedical Named Entity Recognition With No or Limited Number of Examples

The objective of this project was to leverage transfer learning capabilities to develop a method enabling the training of a model to recognize Named Entities for which it has not been trained, with no or minimal supporting training examples. This method necessitates the input data transformation for the binary classification of tokens and employs a transformer architecture for token classification. It is particularly suitable for scenarios where only a few training examples are available, especially in fields where new classes of named entities frequently emerge. The project is the result of cooperation with the pharmaceutical company Bayer.