Jangwon Kim

Amazon

Sr. Applied Scientist

Identifying sections is one of the critical components of understanding medical information from unstructured clinical notes. Most state-of-the-art text classification systems require at least thousands of in-domain text data to achieve high performance. However, collecting in-domain and recent clinical note data with section labels is challenging given the high level of privacy and sensitivity issues. This talk discusses algorithmic ways to leverage large language models for this problem by improving the task transferability of meta-learning-based text classification. This work explores how to make the best use of the source data and presents a task transferability measure named Normalized Negative Conditional Entropy (NNCE).