Probabilistic modeling

By leveraging MTL solutions for negative transfer, we improve the training dynamics of variational autoencoders (VAEs), boosting their performance on heterogeneous datasets.