We developed mixture model inference with discrete-coupled autoencoders (MMIDAS), an unsupervised variational framework that jointly learns discrete clusters and continuous cluster-specific ...