Skip to content

Robustness to rare pathologies

Home / Open Questions / Robustness to rare pathologies

Background: Physiological signal synthesis models often struggle with atypical arrhythmias or pathologies that fall outside the training distribution, as discrete quantization layers may map these irregular patterns to standard codebook entries.

Question / Future Work: There is a need to improve the robustness and generalization of latent manifold quantization models when encountering rare or out-of-distribution pathological cardiac patterns, which currently risk being over-regularized by the discrete codebook.

Why It Matters: Addressing this limitation is critical for the clinical reliability of generative physiological models, ensuring they do not suppress or blur life-saving diagnostic information during synthesis.

Metadata & Links