LongAgent: History-Guided Agentic Search for Longitudinal Outcome Prediction
LongAgent uses prior search results and numerical evidence to choose better feature, time-window, and aggregation combinations for longitudinal medical prediction.
The paper frames longitudinal clinical data as a search problem across heterogeneous variables and uneven time histories. LongAgent autonomously explores candidate representations, using a memory of earlier trials to guide the next ones. On synthetic data, it reports a mean RMSE of 1.7376, narrowly beating the strongest non-agent baseline with statistical significance. On a real clinical dataset, it matches the best baseline rather than clearly surpassing it. ArXiv · AI/CL/LG's note
The paper frames longitudinal clinical data as a search problem across heterogeneous variables and uneven time histories. LongAgent autonomously explores candidate representations, using a memory of earlier trials to guide the next ones. On synthetic data, it reports a mean RMSE of 1.7376, narrowly beating the strongest non-agent baseline with statistical significance. On a real clinical dataset, it matches the best baseline rather than clearly surpassing it. ArXiv · AI/CL/LG's note
score 4