DIYHealth Suite: New AI dataset, model, and benchmark for home health management
A new framework tackles home healthcare with a 900K multimodal dataset and an adaptive AI model.
The DIYHealth Suite, introduced by a team of 12 researchers and accepted at ICML 2026, addresses the shift toward home-based 'Diagnosis-It-Yourself' (DIY) care. The framework tackles three key challenges: heterogeneous home-collected data, the need for models that adapt to evolving individual conditions, and the lack of unified benchmarks. It provides DIYHealth-900K, a large multimodal dataset capturing diverse real-world home care scenarios; DIYHealthGPT, an adaptive foundation model built on a novel Hybrid Hyper Low-Rank Adaptation technique; and DIYHealthBench, the first benchmark for evaluating foundation models on home care tasks.
In extensive experiments, DIYHealthGPT delivered state-of-the-art performance over both general-purpose and medical-specific baselines across 11 home care tasks in open-QA and closed-QA settings. This work lays the groundwork for personalized, accessible health management at home, reducing reliance on hospital-grade devices and enabling continuous monitoring and diagnosis via portable devices and telemedicine.
- DIYHealth-900K: A large-scale multimodal dataset of 900K samples covering real-world home care scenarios.
- DIYHealthGPT uses Hybrid Hyper Low-Rank Adaptation to adapt to changing individual health conditions.
- DIYHealthBench is the first benchmark for evaluating foundation models on 11 home care tasks, outperforming GPT-4 and medical-specific models.
Why It Matters
Enables personalized, accessible health management at home, reducing reliance on expensive hospital equipment.