Efficient multi-task learning with instance selection for biomedical NLP.

Area of Science:

• Biomedical Natural Language Processing (NLP)
• Machine Learning
• Computational Biology

Background:

• Biomedical NLP heavily utilizes large language models and extensive datasets.
• This reliance presents significant computational challenges for researchers and practitioners.

Purpose of the Study:

• To introduce Blue5, an efficient multi-task learning model for biomedical NLP.
• To address computational challenges by incorporating instance selection (IS).

Main Methods:

• Developed Blue5, a multi-task model based on SciFive, integrating instance selection (IS).
• Adapted the E2SC-IS framework for biomedical data, incorporating a calibrated SVM classifier to reduce computational costs.
• Employed multi-task learning (MTL) for efficient processing of biomedical datasets.

Main Results:

• Achieved an average data reduction of 26.6% on the BLUE Benchmark tasks.
• Maintained performance comparable to state-of-the-art models.
• The multi-task SVM configuration proved most effective, showcasing the synergy of IS and MTL.

Conclusions:

• Blue5 offers a practical solution to reduce computational demands in biomedical NLP.
• Enables more scalable and accessible applications of advanced NLP in biomedical research and healthcare.