The research introduces a novel large language model aimed at understanding and reasoning about personal health questions and data. Mobile and wearable devices provide continuous data on an individual’s physiological state and behaviors, which can be used for personal health monitoring and motivation. The model must be able to reason about complex personal health data and produce personalized recommendations. Two papers highlight the capabilities of the model in contextualizing physiological data for personal health tasks and transforming wearable data into personalized health insights.
The model, called Personal Health Large Language Model (PH-LLM), is fine-tuned on expert analysis and self-reported outcomes to provide insights and recommendations related to sleep and fitness patterns. Benchmark datasets are curated to test the model’s ability to produce detailed insights, expert domain knowledge, and predict self-reported assessments of sleep quality. The model performs well in understanding, reasoning, and coaching based on physiological data.
Additionally, an agent framework based on Gemini models is introduced to analyze raw wearable data and provide personalized interpretations and recommendations for health queries. The agent’s capabilities are evaluated on numerical accuracy and reasoning quality, showing improvement over baseline models. The framework can be extended to analyze a broader range of health information.
Overall, the research aims to improve personalized health management by providing accurate and actionable insights from personal health data. The findings represent a step towards AI models that support personalized health recommendations and insights, ultimately helping individuals live longer and healthier lives. The research is a collaborative effort across Google Research, Google Health, Google DeepMind, and partnering teams.
Source link
Source link: http://research.google/blog/advancing-personal-health-and-wellness-insights-with-ai/