Why Parameter Count Is Not Model Complexity: An Introduction to Effective Dimension
Effective dimension quantifies the number of truly independent parameter directions a model can utilize, revealing that a high parameter count does not necessarily imply high model capacity or trainability.