Image Recognition and Reasoning from Long-Tailed Dataset

Learning from datasets with imbalanced distributions is not just a niche concern in real-world machine-learning tasks; it's a pervasive challenge that demands attention and innovative solutions. This thesis echoes this need for deep learning by exploring possible solutions, including the model ensemble, knowledge distillation, and the vision-language foundation model intended to pave the way for future research on topics associated with long-tailed learning.