Erdun Gao

TimeLAVA reveals that a learning-agnostic approach can significantly enhance data valuation in time series, outperforming traditional methods by effectively capturing temporal dynamics.

Aggregating insights from diverse causal discovery experts with LLM-guided reweighting leads to significantly improved causal graph accuracy, even in ambiguous scenarios.

Unseen single-cell perturbation effects can be predicted more accurately by explicitly modeling the latent, dynamic causal processes that drive cellular response.

Why start from scratch when observational data can give you a head start? This paper shows how to design better causal experiments by actively learning *residual* biases, not the whole causal model.