Calibrating deep classifiers with dynamic confidence propagation and adaptive normalization

Although deep neural networks have made significant progress in confidence calibration, conventional methods like temperature scaling and histogram binning face notable limitations in dynamic open-world scenarios due to their reliance on global static parameters or idealized data distribution assumptions. To address these challenges, we propose a Dynamic Confidence Propagation and Alternating Normalization (DCP-AN) framework. Our approach introduces three key innovations: (1) a bidirectional alternating propagation mechanism that enables sample-class confidence synergy through entropy-driven horizontal normalization and KL-divergence-weighted vertical normalization; (2) an adaptive temperature field with dynamic coefficients that achieves differential calibration for non-uniform confidence biases; and (3) a theoretically guaranteed spectral convergence process within 15 iterations. Extensive experiments demonstrate that DCP-AN achieves remarkable improvements: on ImageNet-LT, it boosts tail-class accuracy by 10.3% and reduces expected calibration error by 56%; in cross-domain adaptation, it decreases domain discrepancy by 24% while improving target domain accuracy by 5.5%. Furthermore, DCP-AN maintains computational efficiency with a GPU latency of 1.05 ms and memory overhead under 0.5 MB, making it suitable for real-time deployment.