News

Meditron, EPFL’s new Large Language Model for medical knowledge

EPFL researchers have unveiled Meditron, a groundbreaking open-source Large Language Model (LLM) specifically tailored for the medical field.

EPFL’s innovation is designed to significantly enhance clinical decision-making by providing access to specialized medical knowledge.
EPFL’s innovation is designed to significantly enhance clinical decision-making by providing access to specialized medical knowledge.

Meditron stands apart in the realm of LLMs, which are typically trained on extensive text to learn complex mathematical word relationships. While generalist models like ChatGPT serve a broad spectrum of tasks, Meditron’s medical focus allows it to be more compact yet equally effective. Its development is a leap forward in democratizing access to high-quality medical information, thereby aiding evidence-based clinical decisions.

EPFL’s School of Computer and Communication Sciences developed two versions of Meditron – MEDITRON 7B and 70B, featuring 7 and 70-billion parameters, respectively. Unlike other AI models in the medical domain, which are either closed source or limited in scale, Meditron offers both accessibility and scale. The models were trained on select, high-quality medical data sources, including peer-reviewed literature and diverse clinical practice guidelines, ensuring a broad and accurate knowledge base.

A significant milestone in medical AI development

Lead author Zeming Chen, a doctoral candidate, has highlighted that Meditron surpasses other open-source models in performance and is competitive with leading closed models, signifying a major advance in medical AI. The open-source nature of Meditron is crucial, as it promotes transparency and allows for continuous improvement through contributions from researchers worldwide. This approach is instrumental in fostering trust and reliability in AI advancements. Additionally, the design of Meditron is focused on safety, using transparent sources of high-quality evidence to ensure its efficacy and reliability in clinical settings.

In collaboration with the International Committee of the Red Cross, Meditron also integrates humanitarian clinical practice guidelines, making it sensitive to diverse healthcare contexts. Dr. Javier Elkin of the ICRC praises the tool’s potential in humanitarian settings. A workshop coordinated by the EssentialTech Centre of the EPFL will explore Meditron’s capabilities and challenges, particularly in humanitarian applications.

Furthermore, Meditron aligns with the mission of the new EPFL AI Center, dedicated to leveraging AI for societal benefit. This development is a significant stride in medical technology, offering universal access to medical knowledge and potentially transforming healthcare practices globally.

Handbook for Investors

Our Handbook for Investors provides valuable information about technologies and production costs, taxes and financing, as well as the legal system and infrastructure in Switzerland. Browse through the complete handbook online or download the chapters most relevant to you.

Links

Share

Official program