Enhancing Generative Decoders with Stochastic Training on Biomedical Data with Missingness

Biomedical datasets are often heterogeneous and affected by noise or missing values. Deep Generative Decoders (DGD) provide a promising framework for latent representation learning, but their standard training procedure relies on sample-level stochastic gradient descent (SGD), which performs poorly with incomplete data. To address this, we introduce two stochastic training strategies -- Nested SGD and Feature Dropout -- that incorporate feature-level randomness into optimization. Evaluations on biomedical tabular datasets demonstrate that Nested SGD improves robustness under missingness the best, while Feature Dropout not only improves but also accelerates convergence with lower computational cost. These results suggest that feature-level stochasticity is a practical way to strengthen biomedical AI pipelines.