Your question is Handle Class Imbalance and Noise. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How do you handle severe class imbalance and noisy labels when training a deep learning model on a newly curated dataset?