why 1*1 layer after max-pooling in inception network

Question

I understand that in inception network, 1 * 1 layer is used before 3 * 3 or 5 * 5 filter to do some channel reduction and make computation easier. But why max-pooling then 1