Yang Wang

We boost masked graph autoencoders for molecules by introducing DyCC, a dynamic, chemistry-aware training framework with adaptive masking (GIBMS) and soft label reconstruction (SLG).

We propose ModuLM, a flexible framework for LLM-based molecular relational learning, supporting multimodal inputs and dynamic model construction.

By jointly modeling patient conditions and substructure-aware representations, Our model enhances both accuracy and interpretability, ultimately enabling safer, more personalized drug recommendations.

We enhance long-term time series forecasting by proposing MEW, a new metric, and two modules (IBF and HTM) that help Transformers better use historical data.