When Deep Learning Meets Computer Vision: Challenges, Solutions and Trends

For around a decade now, deep learning has revolutionized computer vision research. However, some key questions remain unanswered:

1. Why do deep-learning networks require extremely high dimensions to encode representational information while human vision does not?
2. Can convolutional neural networks effectively represent image/video data?
3. How can multimodal data fusion be achieved in deep networks?
4. How can machines learn from a few annotated samples?

In this inaugural lecture, I will discuss our team's efforts to address these questions, highlighting progress and future directions at the intersection of deep learning and computer vision.