RecursiveECG: LLM Agent Refines ECG Classifiers Using Failure Evidence
A new framework called RecursiveECG uses an LLM as an offline model designer to refine ECG classifiers by analyzing concrete failures and objective ECG evidence. The system converts curated ECG criteria into validated deterministic functions via Criteria-to-Measurement Compilation, enabling reproducible measurements. Evidence-Grounded Failure Review then analyzes misclassifications to guide iterative improvements. This approach addresses the limitation of aggregate metrics that lack insight into individual case failures. The paper is available on arXiv under ID 2607.24419.
Key facts
- RecursiveECG is an evidence-driven LLM-as-Designer framework
- It refines ECG classifiers based on concrete failures and objective ECG evidence
- Criteria-to-Measurement Compilation converts ECG criteria into validated deterministic functions
- Evidence-Grounded Failure Review analyzes failed and comparable cases
- The approach addresses limitations of aggregate performance metrics
- The paper is on arXiv with ID 2607.24419
- Deep models have advanced 12-lead ECG classification
- LLM agents have potential for automated model design
Entities
Institutions
- arXiv