The field of probabilistic modeling and medical imaging is witnessing significant developments, with a focus on improving the accuracy and generalizability of models. Researchers are exploring novel approaches to learn stochastic functions from partially observed data, such as reformulating forward kernels to depend on inputs and leveraging parameter embedding to integrate physical principles. These advancements have the potential to enhance the performance and reliability of various applications, including image regression, atrial fibrillation ablation, and quantitative MRI synthesis. Noteworthy papers in this area include: Neural Bridge Processes, which proposes a novel method for modeling stochastic functions with dynamic anchors, and SOFA, a deep-learning framework for simulating and optimizing atrial fibrillation ablation. Additionally, A Physics-Driven Neural Network with Parameter Embedding and SynBrain demonstrate the effectiveness of integrating physical principles and probabilistic representation learning in medical imaging and visual-to-fMRI synthesis, respectively.