A Contrastive Deep Learning Approach to Cryptocurrency Portfolio with US Treasuries

To enhance the portfolio performance of cryptocurrency and US treasuries, we introduce a pioneer approach that incorporates contrastive learning to traditional deep learning techniques. The contrastive learning process builds a contrary learning input whenever a market moves, or trader action is performed to enhance the learning supervision. We leverage contrastive learning with both Deep-Q Learning and Proximal Policy Optimization. The result shows promising performance in both downward and upward market environments.