DoctorAgents Refines AutoML for Small Clinical Data.

Clinical machine learning holds immense potential for high-stakes medical decision-making, but its reliable deployment is often hindered by the scarcity, heterogeneity, and temporal complexity of clinical data. Traditional automated machine learning (AutoML) systems struggle with these challenges, relying heavily on brute-force search without explicit reasoning or memory. To address this, researchers have introduced DoctorAgents, an agentic AI framework that redefines AutoML for small clinical datasets. Instead of exhaustive search, DoctorAgents employs a reasoning-driven refinement approach. It autonomously constructs and optimizes entire ML pipelines using specialized large language model (LLM) agents dedicated to generation, validation, and refinement. DoctorAgents utilizes a unique mechanism of backpropagating natural-language feedback through "textual gradient descent" to perform targeted updates, avoiding inefficient brute-force searches. Experiments across diverse clinical tasks demonstrate that DoctorAgents consistently outperforms established AutoML baselines while also generating more interpretable, task-specific representations, which is crucial for medical applications.