Researchers from Shenzhen Research Institute of Big Data and The Chinese University of Hong Kong, Shenzhen, have introduced Apollo, a suite of multilingual medical language models, set to transform the accessibility of medical AI across linguistic boundaries. This is a crucial development in a global healthcare landscape where the availability of medical information in local languages directly impacts the effectiveness of healthcare services.
Medical AI has been evolving at a tremendous pace, with the potential to change healthcare delivery through accurate diagnosis, personalized treatment plans, and unlocking access to extensive medical knowledge. Earlier attempts at developing large language models (LLMs) for medical AI were primarily English-centric and, to a lesser extent, Chinese. This did not account for the vast linguistic diversity of the global medical community, leaving millions that could benefit from medical AI, deprived of the advantages due to language limitations.
Apollo responds to this need, marking a significant step towards inclusive medical AI. The Apollo models were diligently trained using ApolloCorpora, an extensive multilingual dataset, and rigorously assessed against the XMedBench benchmark. This approach enables Apollo to equal or exceed the performance of existing models in languages such as English, Chinese, French, Spanish, Arabic, and Hindi, demonstrating its impressive versatility.
Apollo’s development involved a meticulous methodology. The pre-training corpora were rewritten into a question-and-answer format, and adaptive sampling of training data was used. This created efficient models that could facilitate effective learning transitions, understand and generate multilingual medical information, and augment the capabilities of larger models through a novel proxy tuning technique, thus removing the need for direct fine-tuning.
The models, especially the Apollo-7B, have displayed exceptional performance, setting new benchmarks in multilingual medical LLMs. Apollo opens doors to democratizing medical AI and making several languages compatible with it. Additionally, it broadens the multilingual medical capabilities of larger general LLMs, which is vital for the global adoption of medical AI technologies.
In conclusion, the Apollo project is a promising stride towards democratizing medical AI and making advanced medical knowledge accessible regardless of the linguistic constraints. It tackles the critical gap in global healthcare communication and sets a basis for improved and extensive use of medical AI. Apollo’s successful implementation showcases its capability in bridging the linguistic gap in global healthcare and enhances the applicability of medical AI, thus laying a solid foundation for the future of multilingual medical AI.
