Track new results across pure and applied mathematics.
Search papers, labs, and topics across Lattice.
ECG-RSI recursively refines ECG models by synthesizing verified waveforms and employing feedback-guided intervention search, addressing evolving model weaknesses while maintaining waveform fidelity. This approach significantly boosts Qwen3.5-35B-A3B's performance, raising mean QA accuracy from 34.57% to 56.04% and classification F1 from 24.27 to 34.03 across seven benchmarks over three improvement rounds. The framework decouples error diagnosis from intervention selection and reveals a shift in optimal supervision from direct reinforcement to confusion-focused contrasts.
Recursive self-improvement on ECG models yields substantial accuracy gains by strategically evolving supervision, shifting from direct reinforcement to confusion-based contrasts.
Electrocardiogram (ECG) models need supervision that targets evolving weaknesses and remains faithful to waveform evidence. We introduce ECG-RSI, a recursive self-improvement framework combining verified waveform synthesis with feedback-guided intervention search. Signal remeasurement verifies waveform edits; construction errors guide generator repair, while learner errors motivate interventions compared on a separate selection set. Across seven benchmarks, three rounds raise Qwen3.5-35B-A3B's mean question-answering accuracy from 34.57% to 56.04% and mean classification F1 from 24.27 to 34.03. Across the retained trajectory, the preferred supervision shifts from direct reinforcement to confusion-focused contrasts. Continued feedback correction yields diminishing external gains. These results motivate separating error diagnosis from intervention selection and checking whether gains on the feedback set transfer to independent benchmarks.