Cognitive Decision Correction Improves Automatic Modulation Classification Accuracy
A new arXiv preprint, 2608.02063, presents a post-inference framework aimed at enhancing automatic modulation classification accuracy. This innovative framework employs a cross-fitted residual utility alongside a cognitive decision policy that preserves primary data. Utilizing a structured KAN-Fourier classifier, default probabilities are established, while candidate-specific residual utility is derived from train-split out-of-fold predictions. Results reveal significant accuracy boosts: from 63.632% to 66.332% on RMLA, 65.161% to 66.168% on RMLB, and 77.769% to 79.867% on HISAR. However, controlled comparisons indicate that isolated utility targets do not uniformly improve performance across all datasets.
Key facts
- arXiv preprint 2608.02063 introduces a post-inference framework for automatic modulation classification.
- The framework uses cross-fitted residual utility and a primary-preserving cognitive decision policy.
- A structured KAN-Fourier classifier supplies default probabilities.
- Neural and non-neural candidates provide observable evidence.
- Candidate-specific residual utility is learned from train-split out-of-fold predictions.
- A disjoint validation split freezes action thresholds, approved transitions, conditional routes, and a unified risk mask.
- On RMLA, accuracy improved from 63.632% to 66.332%.
- On RMLB, accuracy improved from 65.161% to 66.168%.
- On HISAR, accuracy improved from 77.769% to 79.867%.
- Controlled comparisons show the isolated utility target does not uniformly benefit all datasets.
Entities
Institutions
- arXiv