![[2602.14901] Picking the Right Specialist: Attentive Neural Process-based Selection of Task-Specialized Models as Tools for Agentic Healthcare Systems](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.14901] Picking the Right Specialist: Attentive Neural Process-based Selection of Task-Specialized Models as Tools for Agentic Healthcare Systems
Summary
The paper presents ToolSelect, a novel method for selecting task-specialized models in healthcare systems, demonstrating superior performance in clinical tasks like diagnosis and report generation.
Why It Matters
This research addresses the challenge of effectively selecting from multiple specialized models in healthcare, which is crucial for improving the accuracy and efficiency of clinical decision-making. By introducing a benchmark and a new selection method, it lays the groundwork for future advancements in agentic healthcare systems.
Key Takeaways
- ToolSelect adapts model selection based on query conditions and model performance.
- The introduction of a benchmark environment for agentic healthcare systems is a significant contribution.
- ToolSelect outperforms 10 state-of-the-art methods across various clinical tasks.
- The research highlights the importance of using multiple specialized models for improved healthcare outcomes.
- The study provides a foundation for future research in task-specialized model selection.
Computer Science > Machine Learning arXiv:2602.14901 (cs) [Submitted on 16 Feb 2026] Title:Picking the Right Specialist: Attentive Neural Process-based Selection of Task-Specialized Models as Tools for Agentic Healthcare Systems Authors:Pramit Saha, Joshua Strong, Mohammad Alsharid, Divyanshu Mishra, J. Alison Noble View a PDF of the paper titled Picking the Right Specialist: Attentive Neural Process-based Selection of Task-Specialized Models as Tools for Agentic Healthcare Systems, by Pramit Saha and 4 other authors View PDF HTML (experimental) Abstract:Task-specialized models form the backbone of agentic healthcare systems, enabling the agents to answer clinical queries across tasks such as disease diagnosis, localization, and report generation. Yet, for a given task, a single "best" model rarely exists. In practice, each task is better served by multiple competing specialist models where different models excel on different data samples. As a result, for any given query, agents must reliably select the right specialist model from a heterogeneous pool of tool candidates. To this end, we introduce ToolSelect, which adaptively learns model selection for tools by minimizing a population risk over sampled specialist tool candidates using a consistent surrogate of the task-conditional selection loss. Concretely, we propose an Attentive Neural Process-based selector conditioned on the query and per-model behavioral summaries to choose among the specialist models. Motivated by t...