Related Concept Videos
Language Development
The critical period for language acquisition suggests that the ability to acquire language is at its peak early in life. As people age, this proficiency decreases. Language development begins very...
Language and Cognition
Improving Translational Accuracy
Improving Translational Accuracy
Components of Language
Cognitive Learning
E. C. Tolman's theory of purposive behavior emphasizes that much behavior is goal-directed. He argued that to understand behavior, we must look at the entire sequence of actions leading to a goal. For instance, high school students study hard, not just due to past reinforcement but also to achieve the goal of getting into a good college.
Tolman introduced the idea that behavior is influenced by...
You might also read
Related Articles
Articles linked to this work by shared authors, journal, and citation graph.
A Web3-Based Patient-Centric Health Data Management System.
Causal Inference on Air Pollutants and Atopic Dermatitis Outpatient Visits in Young Children: An Instrumental Variable Approach.
Design and evaluation of an automated pediatric acute lymphoblastic leukemia registry from clinical data warehouses.
Related Experiment Video
Updated: Jan 13, 2026

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
Beyond Fine-Tuning: Leveraging Domain-Aware In-Context learning with large language models for clinical named entity
Siun Kim1, David Seung U Lee2, Yujin Kim2
1Biomedical Research Institute, Seoul National University Hospital, Seoul, South Korea.
Optimized in-context learning (ICL) with large language models (LLMs) matches or exceeds encoder fine-tuning for clinical named entity recognition (NER). This approach offers efficient adaptation and continuous improvement in healthcare settings without retraining.
Area of Science:
- Natural Language Processing
- Biomedical Informatics
- Machine Learning
Background:
- Clinical named entity recognition (NER) is crucial for organizing unstructured clinical text.
- While large language models (LLMs) offer parameter-free adaptation via in-context learning (ICL), encoder-based fine-tuning has traditionally led in performance for clinical NER.
- This study systematically compares ICL and fine-tuning under realistic conditions to assess their effectiveness.
Purpose of the Study:
- To conduct a systematic comparison of ICL and encoder-based fine-tuning for clinical NER.
- To evaluate the impact of optimizing ICL demonstration selection on performance.
- To determine the viability of ICL in resource-constrained clinical settings.
Main Methods:
- Utilized 2,113 annotated clinical notes from hematologic malignancy patients and 400 MIMIC-IV notes.
- Optimized ICL configurations including instructions, output formats, and demonstration selection strategies using LLaMA-3.3-70B.
- Performed encoder fine-tuning with RoBERTa-large as a baseline, evaluating all models on token-level classification across various scenarios (in-domain, cross-domain, cross-institutional).
Main Results:
- Optimized demonstration selection significantly improved ICL performance, increasing macro F1 by up to 9.4 points.
- In moderate settings, ICL outperformed RoBERTa-large fine-tuning and remained competitive with larger sample pools.
- ICL demonstrated superior data efficiency and achieved substantial gains in cross-institutional transfer without parameter updates, though fine-tuning led at the largest pool size.
Conclusions:
- Optimized domain-aware demonstration selection allows open-source LLM-based ICL to equal or surpass encoder fine-tuning for clinical NER.
- ICL's adaptability and knowledge update capability through demonstration pools, without retraining, are advantageous for dynamic, resource-limited healthcare environments.
- This facilitates continuous improvement in clinical NLP tasks within the healthcare sector.