Advanced Techniques: Fine-Tuning for Medical Data Extraction #

Learning Objectives #

• Understand when and why to fine-tune vs. prompt engineering
• Implement fine-tuning for medical named entity recognition
• Work with medical literature datasets
• Evaluate model performance on healthcare tasks

Fine-Tuning vs. Prompting - When to Use Each? #

• Computational requirements and costs
• Data requirements (how much annotated data do you need?)
• Performance considerations
• Maintenance and updates

Medical Named Entity Recognition Project #

Dataset Introduction:

• Using a subset of PubMed abstracts
• Annotation schemes: drug names, diseases, treatments, dosages
• Data preprocessing for medical text

Interactive Element #

Python Code:

• Base code and step-by-step instructions will be given to students
• Students fine-tune a model on a specific medical extraction task and compare performance with baseline models.

Advanced Data Extraction Tasks #

Briefly share other types of data extraction tasks that are commmonly done in healthcare applications:

• Relationship extraction (drug-disease interactions)
• Extracting clinical trial outcomes
• Identifying contradictory information across papers
• Temporal information extraction (treatment timelines)

Key Considerations for Evaluation and Deployment #

• Evaluation metrics/framework
• Error analysis in healthcare contexts
• Deployment considerations (speed, accuracy, scalability)
• Continuous learning and model updates
• User Trust