This is called the forget gate since the Sigmoid activation function's outputs are very close to \( 0 \) if the argument for the function is very negative, and \( 1 \) if the argument is very positive. Hence we can control the amount of information we want to take from the long-term memory.