The healthcare industry is a prolific producer of data, with sources ranging from clinical trials data such as electronic health record (EHR) text data, health sensor data, medical imaging, patient-generated data and clinical reports to biomolecular data such as drug-molecule data, metabolomics data, nucleic acid sequences, proteins, gene expression data, and single-cell data.

However, much of these complex data are unstructured, incomplete, inconsistent, prone to errors, and heterogeneous, rendering traditional machine learning models less effective in analyzing it. To overcome such, foundation models (FM) such as large language models enable novel applications using such unstructured data, reducing the need for manual feature engineering or large volumes of unlabeled data for pretraining. Foundation models have the potential to revolutionize the clinical understanding, diagnosis, and treatment of complex medical conditions, while also improving the accuracy, and efficiency of healthcare systems. This technology can rapidly enhance patient outcomes, streamline processes, and reduce costs for healthcare providers and patients alike.

Recently, foundation models have been employed in a handful of healthcare tasks such as generating text, creating images from text descriptions, generating captions for images, and utilizing vision-language contrastive learning. Despite early signs of promise, the use of foundation models in healthcare remains largely underexplored and underutilized, limited to a few publicly available medical datasets, which raises concerns about generalization and robustness. Additionally, there is a lack of necessary empirical evidence to support its effectiveness.

Reference

we bold authors that are from FLAME.

[1] Foundation models for generalist medical artificial intelligence, Michael Moor, …, & Pranav Rajpurkar, Nature 2023
[2] Scientific discovery in the age of artificial intelligence, Hanchen Wang …, Yoshua Bengio & Marinka Zitnik, Nature 2023
[3] Large language models generate functional protein sequences across diverse families, Madani, A., ... & Naik, N, Nature Biotechnology 2023
[4] How will generative AI disrupt data science in drug discovery?, Jean-Philippe Vert, Nature Biotechnology 2022
[5] Large Language Models Encode Clinical Knowledge, Karan Singhal, …, Greg S. Corrado, …, Yun Liu, …, arXiv:2212.13138
[6] Towards Expert-Level Medical Question Answering with Large Language Models, K. Singhal, …, Y. Liu, …, G. Corrado, ..., arXiv:2305.09617
[7] Solving quantitative reasoning problems with language models, Aitor Lewkowycz, …, Behnam Neyshabur, …, Vedant Misra, NeurIPS 2022
[8] Training language models with language feedback at scale, J Scheurer, …, A Chen, K Cho, E Perez, arXiv:2303.16755
[9] Health system-scale language models are all-purpose prediction engines, L. Y Jiang, ..., Kyunghyun Cho & Eric K. Oermann , Nature 2023
[10] Transfer learning enables predictions in network biology, Christina V. Theodoris et al., Nature 2023
[11] Using the Veil of Ignorance to align AI systems with principles of justice, Laura Weidinger et al, PNAS 2023
[12] Lessons learned from translating AI from development to deployment in healthcare, Kasumi Widner,, ... Yun Liu, ..., Nature Medicine 2023

The workshop invites researchers to submit working papers in the research areas including, but not limited to:

• Application of foundation models in drug discovery and development
• Predictive modeling for disease diagnosis and prognosis using foundational models
• Natural language processing techniques for clinical text mining and EHR analysis
• Integration of multi-omics data for precision medicine using AI and machine learning
• Explainability and interpretability of foundational models in life sciences and medicine
• Transfer learning and domain adaptation with foundational models
• Ethical considerations and responsible use of foundational models in healthcare
• Benchmarking and evaluation methodologies for foundational models

All submissions are required to use this style file. References and appendices do not count towards the four-page limit, but reviewers will not be required to read beyond the four pages. All accepted papers will be non-archival. Papers have to be anonymized for review, and submitted through the OpenReview portal.