Self-Supervised Representation Learning, Deep Generative Modeling, Cross-Modalities (NLP, CV, Speech)