is overfitting more or less likely when the training set is small or large? Is overfitting more or less

likely when the number of parameters to learn $($such as the number of weights in a neural network$)$

is small or large