In the last decade, deep learning has achieved dramatic success in pattern recognition tasks. However, the characteristic of data hungry training can pose huge challenges in real-world tasks. In addition, the interpretation and domain generalization capability can be vulnerable in existing methods. In this project, we aim to explore to tackle these challenges by developing data-efficient training, interpretable and domain generalizable deep learning techniques.
Students are also encouraged to propose their new ideas within the scope of this topic. Topics of mutual interests can be flexible. Publish high-quality research papers if possible.