# Knowledge Augmentation

At the 2026 World Conference on Traditional Medicine, Yidu Tech released Xiaohuangqi, a traditional Chinese medicine evidence agent that supports classic-text search, guideline and literature tracing, syndrome differentiation assistance, formula lookup and terminology translation. TCM knowledge is complex, terminologically distinctive and spread across many sources, so the system grounds answers in evidence and traces each sentence to its source to reduce the risk of AI hallucination. Its knowledge base draws on People's Medical Publishing House TCM knowledge and more than 8,000 international TCM terms; Yidu ZhiXun and the Yidu clinical assistant are reported in use at more than 40 well-known hospitals nationwide.

At the 2026 World Conference on Traditional Medicine, Yidu Tech released Xiaohuangqi, a traditional Chinese medicine evidence agent that supports classic-text search, guideline and literature tracing, syndrome differentiation assistance, formula lookup and terminology translation. TCM knowledge is complex, terminologically distinctive and spread across many sources, so the system grounds answers in evidence and traces each sentence to its source to reduce the risk of AI hallucination. Its knowledge base draws on People's Medical Publishing House TCM knowledge and more than 8,000 international TCM terms; Yidu ZhiXun and the Yidu clinical assistant are reported in use at more than 40 well-known hospitals nationwide.

Automated feature extraction from weakly annotated bioimaging data performs inconsistently across datasets because models depend on the images alone. A model-agnostic knowledge augmentation method lets machine learning models exploit domain knowledge beyond automated feature extraction, improving classification across seven architectures; combined with masked autoencoder self-supervised learning it raises precision by 2.9% over the next-best method on three in-house microscopy datasets and by 6.8% on average across three public cytomorphology benchmarks.

Automated feature extraction from weakly annotated bioimaging data performs inconsistently across datasets because models depend on the images alone. A model-agnostic knowledge augmentation method lets machine learning models exploit domain knowledge beyond automated feature extraction, improving classification across seven architectures; combined with masked autoencoder self-supervised learning it raises precision by 2.9% over the next-best method on three in-house microscopy datasets and by 6.8% on average across three public cytomorphology benchmarks.