Towards Understanding Overparameterized Neural Networks: Insights Into Imbalances and Neural Representations